In many NLP applications, to mitigate data deficiency in a target task, source data is collected to help with target model training. Existing transfer learning methods either select a subset of source examples that are close to the target domain or try to adapt all source examples into the target domain, then use selected or adapted source examples to train the target model. These methods either incur significant information loss or bear the risk that after adaptation, source examples which are originally already in the target domain may be outside the target domain. To address the limitations of these methods, we propose a four-level optimization based framework which simultaneously selects and adapts source data. Our method can automatically identify in-domain and out-of-domain source examples and apply example-specific processing methods: selection for in-domain examples and adaptation for out-of-domain examples. Experiments on various datasets demonstrate the effectiveness of our proposed method.

Transfer learning (TL) (Zhuang et al., 2020), which aims at improving a model in a target domain by utilizing data from a source domain, has been broadly studied in natural language processing. One paradigm of TL methods (Sun et al., 2011; Song et al., 2012; Wang et al., 2017b; Patel et al., 2018; Qu et al., 2018; Liu et al., 2019b) focuses on selecting a subset of source examples that are close to the target domain and using selected examples as additional training data for the target model. Another paradigm of TL methods (Pan et al., 2010; Ganin et al., 2016; Bousmalis et al., 2017) focuses on adapting the entire set of source examples into the target domain and using adapted source examples as additional training data for the target model. The problem of selection based methods is that unselected examples are discarded. Though having a domain discrepancy with target data, unselected examples still contain useful information that can be leveraged to improve the target model. Discarding them would lead to information loss. The problem of adaptation based methods is that some source examples may already be in the target domain; performing domain adaptation on these source examples is a waste of effort; and, even worse, after adaptation, these source examples may be outside the target domain.

To address the limitations of both paradigms of methods, we propose a new approach which simultaneously performs selection and adaptation of source examples: Our method automatically identifies which source examples are in the same domain as target data (referred to as in-domain source data) and which are not (referred to as out-of-domain source data); for in-domain source data, they are directly used to train the target model; for out-of-domain source data, they are first adapted, then utilized to train the target model. Compared with previous methods, our approach has the following advantage: Instead of using a single way to deal with all source examples (either performing selection or adaptation), our method applies example-specific ways to deal with different examples based on whether they are in-domain or out-of-domain.

Our method is based on four-level optimization, which performs the following four stages end-to-end. At the first stage, a domain distance network is trained based on self-supervised learning. At the second stage, the domain distance network is used to identify out-of-domain source examples and adapt them into the target domain. At the third stage, in-domain source examples selected by the domain distance network and adapted source examples are used to train a target model. At the fourth stage, the trained target model is evaluated on a validation set and data weights of the domain distance network are updated by minimizing the validation losses. Experiments on a variety of datasets demonstrate the effectiveness of our proposed method. To summarize, the major contributions of this work is that we propose a four-level optimization based framework for simultaneous selection and adaptation of source examples in transfer learning and demonstrate the effectiveness of our proposed method on two NLP applications.

Domain Adaptation (DA).

DA (Pan et al., 2010; Ganin et al., 2016; Bousmalis et al., 2017; Sun and Saenko, 2016; Long et al., 2017b; Ben-David et al., 2010; Hoffman et al., 2018; Long et al., 2017a; Kang et al., 2019; Long et al., 2015; Long et al., 2017b; Hoffman et al., 2018; Mitsuzumi et al., 2021) considers the problem of transferring knowledge from a label-rich source domain to a label-deficient target domain where the two domains have distributional discrepancies. There are mainly two paradigms of approaches. One paradigm (Sun and Saenko, 2016; Long et al., 2017b; Ben-David et al., 2010; Kang et al., 2019) is based on metric learning, where a distance metric is defined to measure the distribution discrepancy between domains and domain-invariant representations are learned by minimizing the distance. The other paradigm is based on adversarial learning (Hoffman et al., 2018; Long et al., 2017a; Tzeng et al., 2017; Sankaranarayanan et al., 2018; Motiian et al., 2017), which learns a domain discriminator and a feature learning network adversarially. The domain discriminator is trained to tell whether an instance is from source domain or target domain, while the feature learning network learns domain-invariant features by fooling the domain discriminator. CDAN (Long et al., 2017a) uses multilinear conditioning to capture the cross-covariance between feature representations and classifier predictions, and leverages entropy conditioning to control the uncertainty of classifier predictions. MME (Saito et al., 2019) performs adaptation by maximizing the conditional entropy of unlabeled target data w.r.t the classifier and minimizing it w.r.t the feature encoder. SRDC (Tang et al., 2020) performs deep discriminative clustering with source regularization for unsupervised domain adaptation. These methods perform adaptation on all source data, which leads to waste of efforts and incurs a risk of moving in-domain source data outside the target domain.

Data Selection in Transfer Learning.

Many methods (Jiang and Zhai, 2007; Foster et al., 2010; Moore and Lewis, 2010; Axelrod et al., 2011; Ge and Yu, 2017; Ruder and Plank, 2017; Sivasankaran et al., 2017; Zhang et al., 2017; Guo et al., 2019; Liu et al., 2019a; Tang and Jia, 2019; Wang et al., 2019a, b; Bateson et al., 2020) have been developed for selecting source data that is suitable for training target models, based on reinforcement learning (Patel et al., 2018; Qu et al., 2018; Liu et al., 2019b), adversarial learning (Wang et al., 2019a), curriculum learning (Zhang et al., 2017; Wang et al., 2019b), entropy (Song et al., 2012; Wang et al., 2017b), Bayesian optimization (Ruder and Plank, 2017), multi-task learning (Ge and Yu, 2017), and bi-level optimization (BLO) (Ren et al., 2018, 2020; Hu et al., 2019; Shu et al., 2019; Wang et al., 2020a, b). BLO based approaches select data by minimizing a validation loss, where the lower-level optimization problem trains network weights on a training dataset and the upper-level optimization problem learns data selection variables on a validation set. These methods select part of source data and discard the rest, which incurs information loss.

Transfer Learning (TL).

TL (Pratt, 1993; Mihalkova et al., 2007; Niculescu-Mizil and Caruana, 2007; Pan and Yang, 2009; Luo et al., 2017; Zhuang et al., 2020) aims at training a better target model by utilizing source data. Many TL methods have been developed, including those based on 1) distribution alignment (Huang et al., 2006; Foster et al., 2010; Wang et al., 2017b; Ngiam et al., 2018), 2) regularization (Luo et al., 2008; Tommasi et al., 2010; Duan et al., 2012), 3) adversarial domain-invariant representation learning (Ganin et al., 2016; Long et al., 2017a; Hoffman et al., 2018; Zhang et al., 2019), and 4) latent space projection (Borgwardt et al., 2006; Pan et al., 2010; Long et al., 2013; Wang et al., 2017a). These methods either select part of source data or adapt all source data, which leads to information loss, waste of efforts, and the risk of adapting in-domain source data outside the target domain.

Bi-level Optimization (BLO).

BLO finds broad applications in hyperparameter tuning (Feurer et al., 2015), data selection (Shu et al., 2019; Ren et al., 2020; Wang et al., 2020b), training data generation (Such et al., 2019), neural architecture search (Liu et al., 2018), learning rate adaptation (Baydin et al., 2017), meta learning (Finn et al., 2017), etc. In BLO-based methods, model weights are learned by solving an inner optimization problem and meta parameters are learned by solving an outer optimization problem. The two optimization problems are nested. Different from existing BLO-based methods, our method is based on four-level optimization.

In this section, we present the method for simultaneous selection and adaptation of source examples based on four-level optimization. We aim to train a target model for a specific target domain using dataset Dt. In practical scenarios, the target domain often suffers from a lack of labeled training data. This lack can lead to overfitting on the training data and poor generalization on test data. To address this issue, we utilize a dataset from a source domain, Ds, which has an abundance of labeled examples. However, there is a notable discrepancy between the source and target data. We categorize Ds examples into two types: those that belong to the same domain as Dt (in-domain source data) and those that do not (out-of-domain source data). We aim to select in-domain source data and adapt out-of-domain source data into the target domain, and use selected and adapted source data to train the target model. The overall framework is shown in Figure 1. The notations are shown in Table 1.

Table 1: 

Notations.

NotationMeaning
Dt Target dataset. 
Nt The number of examples in Dt
Dt(tr) Training dataset of Dt
Dt(val) Validation dataset of Dt
dt,i(tr) The i-th example in Dt(tr)
Ds Source dataset. 
Ns The number of examples in Ds
ds, i The i-th example in Ds
q Query example. 
R A set of examples. 
o Binary label regarding whether q and R are from the same domain. 
M The number of SSL training examples. 
W The DDMN’s weight parameters. 
a The weight of an SSL training example. 
A The weights of all SSL training examples. 
W*(AThe optimal solution of W, which depends on A
f(q, Dt;W*(A)) A binary label regarding whether q is out of the target domain. 
W1*(A) The encoder in W*(A). 
W2*(A) The rest of layers in W*(A) besides W1*(A)
z^(q;W1*(A)) The latent representation of q extracted by W1*(A)
z(qThe adapted representation of q
z*(q, W*(A)) The optimal solution of z(q), which depends on W*(A). 
γ A tradeoff parameter in Eq.(2). 
λ A tradeoff parameter in Eq.(6). 
U Target model. 
U^ A sub-network of U, which takes z*(q, W*(A)) as input. 
tgt The target model’s training loss. 
Lt The target model’s training loss on target training data. 
Ls The target model’s training loss on selected source data. 
La The target model’s training loss on adapted source data. 
NotationMeaning
Dt Target dataset. 
Nt The number of examples in Dt
Dt(tr) Training dataset of Dt
Dt(val) Validation dataset of Dt
dt,i(tr) The i-th example in Dt(tr)
Ds Source dataset. 
Ns The number of examples in Ds
ds, i The i-th example in Ds
q Query example. 
R A set of examples. 
o Binary label regarding whether q and R are from the same domain. 
M The number of SSL training examples. 
W The DDMN’s weight parameters. 
a The weight of an SSL training example. 
A The weights of all SSL training examples. 
W*(AThe optimal solution of W, which depends on A
f(q, Dt;W*(A)) A binary label regarding whether q is out of the target domain. 
W1*(A) The encoder in W*(A). 
W2*(A) The rest of layers in W*(A) besides W1*(A)
z^(q;W1*(A)) The latent representation of q extracted by W1*(A)
z(qThe adapted representation of q
z*(q, W*(A)) The optimal solution of z(q), which depends on W*(A). 
γ A tradeoff parameter in Eq.(2). 
λ A tradeoff parameter in Eq.(6). 
U Target model. 
U^ A sub-network of U, which takes z*(q, W*(A)) as input. 
tgt The target model’s training loss. 
Lt The target model’s training loss on target training data. 
Ls The target model’s training loss on selected source data. 
La The target model’s training loss on adapted source data. 
Figure 1: 

The overall framework.

Figure 1: 

The overall framework.

Close modal

3.1 A Four-Level Optimization Framework

We propose a four-level optimization framework to perform simultaneous selection and adaptation of source data. The framework consists of four learning stages which are performed end-to-end.

Stage I.

At the first stage, we learn a domain distance metric network via self-supervised learning (SSL) (He et al., 2019; Chen et al., 2020). This network takes a query example q and a set of examples R as inputs and predicts a binary label representing whether q is in the domain of R. The architecture of this network is as follows. For q and each example in R, an encoder is used to generate a latent representation for the example. Then self-attention (Vaswani et al., 2017) is performed on these latent representations to generate attentive representations. Finally, the attentive representation of q and the averaged attentive representation of examples in R are concatenated and fed into a feedforward layer to predict the binary label.

To learn this domain distance metric network (DDMN), we construct self-supervised training examples. We randomly sample a subset of examples R from Dt, and randomly sample a query example qt from Dt and a query example qs from Ds. We label (qt, R) as a positive pair since qt and R are both from target domain, and label (qs, R) as a negative pair since qs is from source domain and R is from the target domain. This procedure repeats M times, yielding 2M training examples denoted by Dssl={(qi,Ri,oi)}i=12M where oi is a binary label representing whether qi and Ri are from the same domain. Let W denote weight parameters of the DDMN. We learn W by minimizing the binary classification loss: i=12M(qi,Ri,oi;W) where is a cross-entropy loss.

Note that these self-supervised examples may be noisy since the binary labels are given based on heuristics without human scrutiny. It could be the case that qs happens to be in the same domain as R while qt is not. It is necessary to automatically identify and remove these noisy examples. To achieve this goal, we associate each example (qi, Ri, oi) with a weight ai ∈ [0,1]. The smaller ai is, the more likely that (qi, Ri, oi) is noisy. We multiply ai to the loss (qi, Ri, oi;W). If ai is close to zero, ai(qi, Ri, oi;W) is made close to zero, which effectively removes (qi, Ri, oi) from the training set. This stage amounts to solving the following optimization problem:
(1)
where A={ai}i=12M. W*(A) denotes that W* is a function of A. This is because W* is a function of the loss which is a function of A. A is tentatively fixed at this stage and will be updated at a later stage. A cannot be updated at this stage. Otherwise, all the values in A would be zero.

Stage II.

At the second stage, we use the learned domain distance network to identify out-of-domain source examples and adapt them into the target domain. For each data example q from the source dataset Ds, we feed q and the target dataset Dt into W*(A), which predicts a binary label f(q, Dt;W*(A)). If f(q, Dt;W*(A)) = 0, it means q is out of the target domain, and we use the domain distance network to adapt it into the target domain. The adaptation is performed in the following way. In W*(A), let W1*(A) denote the encoder (containing all layers before self-attention) and W2*(A) denote the rest of layers (including self-attention and feedforward layers) used for predicting the binary label. Let z^(q;W1*(A)) denote the latent representation of q extracted by W1*(A)—the layer before self-attention. We learn another representation z(q) of q which falls into the target domain (in the latent space) and is close to z^(q;W1*(A)). Closeness is measured using L2 distance. To encourage z(q) to fall into the target domain, we encourage W2*(A) to predict that z(q) is in the target domain, i.e., minimizing the binary classification loss l(z(q),Dt,tq=1;W2*(A)). At this stage, we solve the following optimization problem for each source example q that is predicted to be out-of-domain:
(2)
where γ is a tradeoff parameter.

Stage III.

At the third stage, we use selected and adapted source examples, together with the target training data Dt(tr), to train the target model U. Let tgt denote the loss function of the target task, Ns and Nt denote the number of examples in Ds and Dt, and f(q, Dt;W*(A)) denote the binary label predicted by the domain distance network W*(A) regarding whether a source example q is in the target domain. dt,i(tr) denotes the i-th example in Dt(tr), ds, i denotes the i-th example in Ds. For a source example which is predicted to be in the target domain (where f(ds, i, Dt;W*(A)) = 1), it is directly used to train the target model U by minimizing the loss tgt(U, ds, i). For a source example which is predicted to be out of the target domain (where f(ds, i, Dt;W*(A)) = 0), its adapted representation z*(ds, i, W*(A)) obtained in the second stage is used to train the target model. Note that since z*(ds, i, W*(A)) is already a latent representation, only part of weight parameters (denoted by U^) in the target model U is needed to make prediction on z*(ds, i, W*(A)). We define the loss on target training data as:
(3)
the loss on selected source data as:
(4)
and the loss on adapted data as:
(5)
At this stage, we solve the following optimization problem:
(6)
where λ is a tradeoff parameter.

Stage IV.

At the fourth stage, we use the trained target model to make predictions on the validation dataset Dt(val) of the target task. We update the weights A of self-supervised training examples in the first stage by minimizing the validation loss:
(7)

A Four-level Optimization Based Framework.

Putting all these pieces together, we have the following four-level optimization based framework.
(8)
To make the objective at the third stage differentiable, we approximate f(ds, i, Dt;W*(A)) using the probability calculated by W*(A) regarding whether ds, i and Dt are in the same domain.

3.2 Optimization Algorithm

We leverage a gradient-based method (Liu et al., 2018) to solve the problem in Eq.(8). Convergence of this algorithm has been analyzed (Ghadimi and Wang, 2018; Grazzi et al., 2020; Ji et al., 2021; Liu et al., 2021; Yang et al., 2021). At each level of the optimization problem, the exact value of the optimal solution (on the left-hand side of the equal sign, marked with *) is computationally expensive to compute. To address this problem, following Liu et al. (2018), we approximate the optimal solution using a one-step gradient descent update and plug the approximation into the next level of optimization problem. In the sequel, ·· denotes partial derivative. d·d· denotes an ordinary derivative. Following Liu et al. (2018), we approximate W*(A) using one-step gradient descent update of W:
(9)
We plug W*(A) ≈ W′ into the loss function at the second stage and get an approximated objective. Let W2′ and W1′ denote the approximations of W2*(A) and W1*(A), respectively. The approximated objective is O(z(q),W2,W1)=l(z(q),Dt,tq=1;W2)+γz(q)z^(q;W1)22. We approximate z*(q, W*(A)) using one-step gradient descent update of z(q) w.r.t the approximated objective:
(10)
We plug z*(q, W*(A)) ≈ z′(q) and W*(A) ≈ W′ into the objective at the third stage and get an approximated objective. Let g(ds, i, Dt;W*(A)) denote the probability that ds, i and Dt are in the same domain. We approximate U*(W*(A)) using one-step gradient descent update of U w.r.t the approximated objective:
(11)
Finally, we plug the approximation U*(W*(A)) ≈ U′ into the validation loss at the fourth stage and update A by minimizing the approximated loss using gradient descent:
(12)
For AL(U,Dt(val)), it can be computed as:
(13)
where
(14)
(15)
(16)
(17)
The gradient descent update of A in Eq.(13) can run one or more steps. After A is updated, the one-step gradient-descent approximations in Eq.(9), (10), and (11), which are functions of A, change with A and need to be re-updated. Then, the gradient of A, which is a function of one-step gradient-descent approximations, needs to be re-calculated and is used to refresh A. In sum, the update of A and the updates of one-step gradient-descent approximations mutually depend on each other. These updates are performed iteratively until convergence. Algorithm 1 shows the algorithm.

graphic

In the gradient of A calculated using the chain rule, the number of chains is the same as the number of levels in our proposed four-level optimization formulation. This shows that this optimization algorithm preserves the four-level nested optimization nature of the proposed formulation.

3.3 Reduce Computation and Memory Costs

To reduce computation and memory costs, we adopt the following methods.

  • We reduced the frequencies of updating (including calculating hypergradients of) the weights A of self-supervised training examples. They were updated every 8 mini-batches (i.e., iterations) instead of on every mini-batch. We empirically found that this greatly reduced computational costs without significantly sacrificing accuracy. The rest of the parameters were updated on every mini-batch.

  • We added a decorrelation regularizer (Cogswell et al., 2015) on W and U, which significantly speeds up convergence and allows reducing the number of epochs by half without sacrificing convergence quality.

  • Parameter tying was performed to reduce the number of weight parameters and computation costs. We let W and U share the same feature learning layers. These layers account for >95% of parameters in each of these models. Sharing them across models significantly reduces the number of total parameters, which reduces the computational costs of updating these parameters.

  • We optimized the implementation of our method to speed up computation by leveraging techniques including 1) automatic mixed precision (Micikevicius et al., 2017), 2) using multiple (4, specifically) workers and pinned memory in PyTorch DataLoader, 3) using cudNN autotuner, 4) kernel fusion, and so forth.

3.4 Applications

In this section, we apply the proposed four-level optimization framework for two NLP applications.

Text Classification.

In many text classification problems, training data in a target domain is limited. To address the lack of target training data, one can leverage data from a source domain.

Visual Question Answering on Pathology Images.

Pathology imaging (Mohan, 2015) is broadly used for identifying the causes and effects of diseases or injuries. Given a pathology image, being able to answer questions about the clinical findings contained in the image is very important for medical decision-making (He et al., 2020). However, collecting a large-scale visual question answering (VQA) dataset is challenging, due to the lack of doctors for making questions and answers from pathology images. A dataset collected in He et al. (2020) has about 33K question-answer pairs generated from around 5K pathology images. Although largest of its kind, it is still relatively small compared with common VQA datasets. To mitigate the deficiency of training data, we collect an auxiliary source dataset. From the pathology literature, we collect 1792 pathology figures and create 36,471 VQA questions using the method proposed in He et al. (2020).

In this section, we present experimental results on text classification and visual question answering on pathology images. Following the common data assumption in transfer learning, the amount of labeled source data in our experiments is significantly larger than that of target data. Every experiment runs 4 times with different random initializations. For all experiments, we performed significance tests using double-sided t-tests. The p-values of our methods against baselines are all less than 0.001, which shows that our methods are significantly better than baselines.

4.1 Text Classification

Dataset.

Following Gururangan et al. (2020), we experiment with four domains: biomedical, computer science, news, and reviews. For the biomedical domain, we use two target datasets: ChemProt (Kringelum et al., 2016) and RCT (Dernoncourt and Lee, 2017), and one source dataset which contains 2.68 million full-text papers from S2ORC (Lo et al., 2019) with 7.55 billion tokens. For the computer science domain, we use two target datasets: ACL-ARC (Jurgens et al., 2018) and SciERC (Luan et al., 2018), and one source dataset which contains 2.22 millon full-text papers from S2ORC (Lo et al., 2019) with 8.1 billion tokens. For the news domain, we use two target datasets: HyperPartisan (Kiesel et al., 2019) and AGNews (Zhang et al., 2015), and one source dataset which contains 11.9 million articles from RealNews (Zellers et al., 2019) with 6.66 billion tokens. For the reviews domain, we use two target datasets: Helpfulness (McAuley et al., 2015) and IMDB (Maas et al., 2011), and one source dataset which contains 24.75 million articles from Amazon Reviews (He and McAuley, 2016) with 2.11 billion tokens. Statistics of the target datasets are summarized in Table 2. In our method, we split the original target training set into a new training set and a validation set, with a ratio of 1:1. The new training set is used as Dcls(tr) and the validation set is used as Dcls(val). Note that baseline methods are trained on the combination of Dcls(tr) and Dcls(val).

Table 2: 

Statistics of datasets used in Gururangan et al. (2020).

DomainDatasetLabel TypeTrainDevTestClasses
Biomedical ChemProt relation classification 4169 2427 3469 13 
RCT abstract sent. roles 180040 30212 30135 
Computer ACL-ARC citation intent 1688 114 139 
Science SciERC relation classification 3219 455 974 
News HyperPartisan partisanship 515 65 65 
AGNews topic 115000 5000 7600 
Reviews Helpfulness review helpfulness 115251 5000 25000 
IMDB review sentiment 20000 5000 25000 
DomainDatasetLabel TypeTrainDevTestClasses
Biomedical ChemProt relation classification 4169 2427 3469 13 
RCT abstract sent. roles 180040 30212 30135 
Computer ACL-ARC citation intent 1688 114 139 
Science SciERC relation classification 3219 455 974 
News HyperPartisan partisanship 515 65 65 
AGNews topic 115000 5000 7600 
Reviews Helpfulness review helpfulness 115251 5000 25000 
IMDB review sentiment 20000 5000 25000 

Baselines.

We compare our method with the following baselines. In baseline methods, the target data used for training the target model includes both training and validation sets.

  • Domain adaptive pretraining (DAPT) (Gururangan et al., 2020): Given an RoBERTa-base model which has been pretrained on large amounts of corpora (used in Liu et al., 2019c), we continue to pretrain it on source data, then finetune it on target data.

  • Task adaptive pretraining (TAPT) (Gururangan et al., 2020): Given an RoBERTa-base model which has been pretrained on large amounts of corpora (used in Liu et al., 2019c), we continue to pretrain it on input texts of each target dataset.

  • Data selection methods for transfer learning, based on Bayesian optimization (BO) (Ruder and Plank, 2017), minimax game (MMG) (Wang et al., 2019a), learning to select instance (LSI) (Huan et al., 2021).

  • Domain adaptation methods, including DANN (Ganin et al., 2016), CDAN (Long et al., 2017a), MME (Saito et al., 2019), SRDC (Tang et al., 2020), SSDA (Kim and Kim, 2020), GDA (Mitsuzumi et al., 2021), and ATDOC (Liang et al., 2021).

  • SimCSE (Gao et al., 2021): A contrastive learning method. The same input sentence is fed into a pretrained RoBERTa encoder twice by applying different dropout masks, to get two different embeddings. These two embeddings are labeled as being “similar”. Embeddings of different sentences are labeled as being “dissimilar”. Contrastive learning is performed on these “similar” and “dissimilar” pairs.

Implementation Details.

In the domain distance network, the hyperparameters of self-attention and feed-forward layer are the same as those in Transformer (Vaswani et al., 2017). The subset R has varying cardinality (sampled uniformly). M is set to 10k. The tradeoff parameter γ and λ is set to 0.1 and 0.5 respectively. Baselines and our method receive a similar amount of tuning time and efforts. F1 is used as the evaluation metric. We use RoBERTa-base as a data encoder. For a fair comparison, most of our hyperparameters are the same as those in Gururangan et al. (2020). The maximum text length was set to 512. For all datasets, we used a batch size of 16 with gradient accumulation. We used the AdamW optimizer (Loshchilov and Hutter, 2017) with a warm-up proportion of 0.06, a weight decay of 0.1, and an epsilon of 1e-6. In AdamW, β1 and β2 are set to 0.9 and 0.98, respectively. The maximum learning rate was 2e-5. For the reader’s convenience, we summarize the hyperparameters as follows.

  • The number M of self-supervised training examples: 10k

  • Tradeoff parameters γ and λ: 0.1, 0.5

  • Maximum text length: 512

  • Batch size: 16

  • Optimizer: AdamW

  • Warm-up proportion, weight decay, and epsilon in AdamW: 0.06, 0.1, and 1e-6

  • β1 and β2 in AdamW: 0.9, 0.98

  • Maximum learning rate: 2e-5

To tune the hyperparameters, we randomly split the validation set into two equal-sized subsets, denoted by A and B. For each configuration of hyperparameters, we use the validation set A to learn the importance weights of self-supervised training examples. Then we measure the performance of the trained model on validation set B. Hyperparameter values yielding the best performance on validation set B are selected. Each baseline method received an equal amount of tuning effort as that for our method.

Results and Analysis.

Table 3 shows the results. From this table, we make the following observations. First, our method outperforms source data selection methods including BO, MMG, and LSI. In these baseline methods, out-of-domain source examples are discarded, which incurs information loss. In contrast, our method adapts out-of-domain source examples into the target domain and uses the adapted examples to train the target model. Second, our method works better than domain adaptation methods including DANN, CDAN, MME, SRDC, SSDA, GDA, and ATDOC. The reason is: These methods perform adaptation on all source examples without identifying which ones are already in the target domain; as a result, some in-domain source examples may be adapted out of the target domain. In contrast, our method first identifies which source examples are already in-domain and only performs adaptation on out-of-domain examples. Third, our method outperforms vanilla RoBERTa-base which does not leverage source data to learn representations. This demonstrates that leveraging source data is helpful for improving the target model. Fourth, our method performs better than DAPT. In DAPT, all source examples are leveraged to pretrain the target encoder without considering the fact that some source examples have large domain discrepancy with the target domain and are not suitable for pretraining the target encoder. Fifth, our method outperforms TAPT and SimCSE. These methods do not leverage auxiliary source data or select/adapt source data.

Table 3: 

Text classification results. Following Gururangan et al. (2020), the results are micro-F1 for ChemProt and RCT, and macro-F1 for other datasets. For each xy entry, x and y represent the mean and standard deviation of four random runs, respectively.

MethodChemProtRCTACL-ARCSciERCHyperPartisanAGNewsHelpfulnessIMDBAverage
RoBERTa-base 81.91.0 87.20.1 63.05.8 77.31.9 86.60.9 93.90.2 65.13.4 95.00.2 81.3 
TAPT 82.60.4 87.70.1 67.41.8 79.31.5 90.45.2 94.50.1 68.51.9 95.50.1 83.2 
DAPT 84.20.2 87.60.1 75.42.5 80.81.5 88.25.9 93.90.2 66.51.4 95.40.1 84.0 
SimCSE 83.20.2 87.60.1 69.52.6 80.50.7 90.92.7 94.70.1 68.71.7 95.70.1 83.9 
BO 83.30.2 87.51.0 75.43.3 79.11.3 88.14.2 94.00.2 68.42.1 95.30.2 83.9 
MMG 83.00.4 87.41.0 75.14.1 79.60.8 87.92.1 94.20.1 66.71.5 95.70.1 83.7 
LSI 83.20.3 87.51.0 75.34.5 80.20.6 88.71.9 94.50.1 68.61.0 95.40.1 84.2 
DANN 83.50.3 87.70.1 75.53.7 80.51.1 90.43.0 94.30.1 66.82.8 95.20.1 84.2 
CDAN 83.80.2 87.40.1 75.92.4 80.91.4 88.75.5 94.10.2 67.33.4 95.80.2 84.2 
MME 84.00.1 87.40.1 75.75.1 80.50.8 89.22.8 94.60.2 68.92.7 95.40.1 84.5 
SRDC 84.30.3 87.30.1 75.73.4 80.71.0 88.93.5 94.10.1 67.63.0 95.10.1 84.2 
SSDA 83.90.5 87.90.1 75.92.7 81.01.4 88.52.6 94.40.1 67.01.6 95.70.2 84.3 
GDA 84.10.3 87.30.1 75.42.4 80.80.9 90.54.1 94.20.2 66.82.4 95.50.1 84.3 
ATDOC 84.50.2 87.50.1 75.63.6 81.11.2 89.05.7 93.90.2 67.42.8 95.10.1 84.3 
 
No-Adapt 84.10.2 87.30.1 75.92.2 81.41.0 90.34.2 94.10.1 66.93.3 95.30.2 84.4 
No-In-Domain 84.50.1 87.50.1 75.54.5 81.61.7 88.83.9 94.60.2 67.13.1 95.60.1 84.4 
Separate 85.30.5 88.10.1 75.82.9 82.70.9 90.13.6 94.00.2 68.31.5 95.20.1 84.9 
MMD 85.90.4 88.30.1 75.54.9 82.31.1 89.51.8 94.10.1 68.91.2 95.40.2 85.0 
AD 85.60.2 88.70.1 75.93.1 82.60.8 90.33.0 94.70.2 67.41.7 95.80.1 85.1 
WMVL 85.50.4 88.50.1 75.72.8 81.91.3 90.64.6 94.60.1 68.12.3 95.10.1 85.0 
H-Divergence 85.40.3 88.70.1 75.94.6 82.10.9 89.21.9 94.20.1 68.52.6 95.30.1 84.9 
 
Our full method 87.10.2 90.40.1 77.62.3 84.40.7 92.41.2 95.70.1 70.90.8 96.60.1 86.9 
MethodChemProtRCTACL-ARCSciERCHyperPartisanAGNewsHelpfulnessIMDBAverage
RoBERTa-base 81.91.0 87.20.1 63.05.8 77.31.9 86.60.9 93.90.2 65.13.4 95.00.2 81.3 
TAPT 82.60.4 87.70.1 67.41.8 79.31.5 90.45.2 94.50.1 68.51.9 95.50.1 83.2 
DAPT 84.20.2 87.60.1 75.42.5 80.81.5 88.25.9 93.90.2 66.51.4 95.40.1 84.0 
SimCSE 83.20.2 87.60.1 69.52.6 80.50.7 90.92.7 94.70.1 68.71.7 95.70.1 83.9 
BO 83.30.2 87.51.0 75.43.3 79.11.3 88.14.2 94.00.2 68.42.1 95.30.2 83.9 
MMG 83.00.4 87.41.0 75.14.1 79.60.8 87.92.1 94.20.1 66.71.5 95.70.1 83.7 
LSI 83.20.3 87.51.0 75.34.5 80.20.6 88.71.9 94.50.1 68.61.0 95.40.1 84.2 
DANN 83.50.3 87.70.1 75.53.7 80.51.1 90.43.0 94.30.1 66.82.8 95.20.1 84.2 
CDAN 83.80.2 87.40.1 75.92.4 80.91.4 88.75.5 94.10.2 67.33.4 95.80.2 84.2 
MME 84.00.1 87.40.1 75.75.1 80.50.8 89.22.8 94.60.2 68.92.7 95.40.1 84.5 
SRDC 84.30.3 87.30.1 75.73.4 80.71.0 88.93.5 94.10.1 67.63.0 95.10.1 84.2 
SSDA 83.90.5 87.90.1 75.92.7 81.01.4 88.52.6 94.40.1 67.01.6 95.70.2 84.3 
GDA 84.10.3 87.30.1 75.42.4 80.80.9 90.54.1 94.20.2 66.82.4 95.50.1 84.3 
ATDOC 84.50.2 87.50.1 75.63.6 81.11.2 89.05.7 93.90.2 67.42.8 95.10.1 84.3 
 
No-Adapt 84.10.2 87.30.1 75.92.2 81.41.0 90.34.2 94.10.1 66.93.3 95.30.2 84.4 
No-In-Domain 84.50.1 87.50.1 75.54.5 81.61.7 88.83.9 94.60.2 67.13.1 95.60.1 84.4 
Separate 85.30.5 88.10.1 75.82.9 82.70.9 90.13.6 94.00.2 68.31.5 95.20.1 84.9 
MMD 85.90.4 88.30.1 75.54.9 82.31.1 89.51.8 94.10.1 68.91.2 95.40.2 85.0 
AD 85.60.2 88.70.1 75.93.1 82.60.8 90.33.0 94.70.2 67.41.7 95.80.1 85.1 
WMVL 85.50.4 88.50.1 75.72.8 81.91.3 90.64.6 94.60.1 68.12.3 95.10.1 85.0 
H-Divergence 85.40.3 88.70.1 75.94.6 82.10.9 89.21.9 94.20.1 68.52.6 95.30.1 84.9 
 
Our full method 87.10.2 90.40.1 77.62.3 84.40.7 92.41.2 95.70.1 70.90.8 96.60.1 86.9 

4.2 Visual Question Answering on Pathology Images

Datasets.

For the target dataset, we use PathVQA (He et al., 2020), which contains 1,670 pathology images and 32,795 question-answer pairs. Of these, 16,466 questions are open-ended, with the following types: what, where, when, whose, how, why, how much/how many. The rest are close-ended “yes/no” questions. Based on images, the dataset is split into a training, validation, and test set with a ratio of 3:1:1 approximately. Note that baseline methods are trained on the combination of the training and validation sets. We collected a source dataset containing 1792 pathology figures extracted from papers in medRxiv where each figure has a caption. We create 36,471 VQA questions using the method proposed in He et al. (2020). This source dataset will be made available publicly.

Implementation Details.

For the target model, we experimented with two state-of-the-art VQA models—LXMERT (Tan and Bansal, 2019) and bilinear attention networks (BAN) (Kim et al., 2018)—each containing an image encoder, a question encoder, and an answer generation head. Hyperparameters mostly follow those in previous work (Tan and Bansal, 2019; Kim et al., 2018; Yang et al., 2016). For LXMERT, the hidden size of the text encoder was set to 768.

The initial learning rate was set to 5e-5 with the Adam (Kingma and Ba, 2014) optimizer used. The batch size was set to 256. The model was trained for 200 epochs. For BAN, words in questions and answers were represented using GloVe (Pennington et al., 2014) vectors. The initial learning rate was set to 0.005 with the Adamax optimizer (Kingma and Ba, 2014) used. The batch size was set to 512. The model was trained for 200 epochs.

From questions and answers in the PathVQA dataset, we create a vocabulary of 4,631 words that have the highest frequencies. Data augmentation is applied to images, including shifting, scaling, and shearing. We compare with baselines similar to those in Section 4.1. The Pretrain baseline works as follows: We pretrain the target encoder using source data, then finetune the target encoder using target data. Three metrics were used for evaluation, including BLEU (Papineni et al., 2002), macro-averaged F1 (Goutte and Gaussier, 2005), and accuracy (Malinowski and Fritz, 2014). We implement the methods using PyTorch and perform training on four GTX 1080Ti GPUs.

Results and Analysis.

The results are shown in Table 4. From this table, we make similar observations as those in Table 3. The analysis of reasons is similar to that for results in Table 3. The training time of our method is similar to that of baselines. Figure 2 shows some randomly sampled source pathology figures identified by our method as being out of the target domain. These images contain subfigures and texts, which are different from the target dataset.

Table 4: 

Results on the PathVQA dataset. Runtime (hours) for training is measured on a 1080TI GPU.

Accuracy(%)BLEU-1(%)BLEU-2(%)BLEU-3(%)F1(%)AverageRuntime(h)
LXMERT (Tan and Bansal, 2019) based experiments 
Vanilla LXMERT (Tan and Bansal, 201957.6 57.4 3.1 1.3 9.9 25.9 29 
Pretrain (He et al., 201659.3 59.0 4.6 2.6 11.3 27.4 35 
BO (Ruder and Plank, 201759.5 58.9 3.8 2.7 11.4 27.3 41 
MMG (Wang et al., 2019a59.3 58.4 3.7 2.6 10.9 27.0 38 
LSI (Huan et al., 202159.2 58.8 3.5 2.8 10.6 27.0 32 
DANN (Ganin et al., 201659.9 59.4 4.3 3.2 11.2 27.6 30 
CDAN (Long et al., 2017a60.2 58.9 4.7 3.3 11.9 27.8 35 
MME (Saito et al., 201959.7 59.0 3.6 2.9 11.7 27.4 30 
SRDC (Tang et al., 202058.8 58.7 4.8 2.9 12.2 27.5 39 
SSDA (Kim and Kim, 202059.3 58.3 4.4 3.0 12.0 27.4 34 
GDA (Mitsuzumi et al., 202159.8 59.1 4.6 2.8 12.3 27.7 36 
ATDOC (Liang et al., 202159.6 59.6 4.3 3.1 12.2 27.8 30 
Ours 62.5 61.1 5.2 3.7 12.9 29.1 29 
 
BAN (Kim et al., 2018) based experiments 
Vanilla BAN (Kim et al., 201855.1 56.2 3.2 1.2 8.4 24.8 25 
Pretrain (He et al., 201658.4 58.6 4.3 1.6 10.3 26.6 28 
BO (Ruder and Plank, 201758.3 58.5 4.2 2.0 10.9 26.8 32 
MMG (Wang et al., 2019a58.7 58.1 4.6 2.3 11.2 27.0 29 
LSI (Huan et al., 202158.3 58.7 4.3 2.1 11.5 27.0 30 
DANN (Ganin et al., 201658.8 58.4 4.3 2.3 11.0 27.0 31 
CDAN (Long et al., 2017a58.6 58.6 4.5 2.5 11.4 27.1 33 
MME (Saito et al., 201959.1 58.9 4.9 2.3 11.1 27.3 26 
SRDC (Tang et al., 202058.9 58.7 4.7 2.5 11.6 27.3 35 
SSDA (Kim and Kim, 202059.3 59.0 4.4 2.6 11.4 27.3 32 
GDA (Mitsuzumi et al., 202159.5 58.8 4.9 2.4 11.7 27.5 30 
ATDOC (Liang et al., 202159.1 59.1 5.0 2.7 11.5 27.5 29 
Ours 62.4 61.7 5.6 3.2 12.5 29.1 26 
Accuracy(%)BLEU-1(%)BLEU-2(%)BLEU-3(%)F1(%)AverageRuntime(h)
LXMERT (Tan and Bansal, 2019) based experiments 
Vanilla LXMERT (Tan and Bansal, 201957.6 57.4 3.1 1.3 9.9 25.9 29 
Pretrain (He et al., 201659.3 59.0 4.6 2.6 11.3 27.4 35 
BO (Ruder and Plank, 201759.5 58.9 3.8 2.7 11.4 27.3 41 
MMG (Wang et al., 2019a59.3 58.4 3.7 2.6 10.9 27.0 38 
LSI (Huan et al., 202159.2 58.8 3.5 2.8 10.6 27.0 32 
DANN (Ganin et al., 201659.9 59.4 4.3 3.2 11.2 27.6 30 
CDAN (Long et al., 2017a60.2 58.9 4.7 3.3 11.9 27.8 35 
MME (Saito et al., 201959.7 59.0 3.6 2.9 11.7 27.4 30 
SRDC (Tang et al., 202058.8 58.7 4.8 2.9 12.2 27.5 39 
SSDA (Kim and Kim, 202059.3 58.3 4.4 3.0 12.0 27.4 34 
GDA (Mitsuzumi et al., 202159.8 59.1 4.6 2.8 12.3 27.7 36 
ATDOC (Liang et al., 202159.6 59.6 4.3 3.1 12.2 27.8 30 
Ours 62.5 61.1 5.2 3.7 12.9 29.1 29 
 
BAN (Kim et al., 2018) based experiments 
Vanilla BAN (Kim et al., 201855.1 56.2 3.2 1.2 8.4 24.8 25 
Pretrain (He et al., 201658.4 58.6 4.3 1.6 10.3 26.6 28 
BO (Ruder and Plank, 201758.3 58.5 4.2 2.0 10.9 26.8 32 
MMG (Wang et al., 2019a58.7 58.1 4.6 2.3 11.2 27.0 29 
LSI (Huan et al., 202158.3 58.7 4.3 2.1 11.5 27.0 30 
DANN (Ganin et al., 201658.8 58.4 4.3 2.3 11.0 27.0 31 
CDAN (Long et al., 2017a58.6 58.6 4.5 2.5 11.4 27.1 33 
MME (Saito et al., 201959.1 58.9 4.9 2.3 11.1 27.3 26 
SRDC (Tang et al., 202058.9 58.7 4.7 2.5 11.6 27.3 35 
SSDA (Kim and Kim, 202059.3 59.0 4.4 2.6 11.4 27.3 32 
GDA (Mitsuzumi et al., 202159.5 58.8 4.9 2.4 11.7 27.5 30 
ATDOC (Liang et al., 202159.1 59.1 5.0 2.7 11.5 27.5 29 
Ours 62.4 61.7 5.6 3.2 12.5 29.1 26 
Figure 2: 

Randomly sampled source pathology figures identified by our method as being out of the target domain.

Figure 2: 

Randomly sampled source pathology figures identified by our method as being out of the target domain.

Close modal

4.3 Ablation Studies

4.3.1 Ablation by Removing Certain Components

To better understand the contributions of individual components in our framework, we perform the following ablation studies.

  • No-Adapt. Out-of-domain source examples are discarded instead of being adapted. This is equivalent to removing stage II and the loss term I(f(ds,i,Dt;W*(A))=0)tgt(U^,z*(ds,i,W*(A))) in Eq.(6) of stage III.

  • No-In-Domain. In-domain source examples are discarded instead of being used for training the target model. This is equivalent to removing the loss term I(f(ds,i,Dt;W*(A))=1)tgt(U,ds,i) in Eq.(6) of stage III.

  • Separate. Different stages are performed separately instead of jointly.

Table 3 shows ablation study results. From this table, we make the following observations. First, our full method works better than No-Adapt. This shows that it is beneficial to adapt out-of-domain source examples into the target domain and use adapted examples to train the target model, and our method is effective in achieving this goal. Second, our full method outperforms No-In-Domain, which demonstrates that the source examples selected by our method are useful for training the target model. Third, our full method achieves better performance than Separate. Our full method performs source data selection, adaptation, and target model training jointly, which enables these different tasks to mutually influence each other to achieve the best overall performance. Such a mechanism is lacking in Separate.

4.3.2 Ablation on the Adaptation Component

We perform an ablation study of the adaptation component in stage II of our framework by replacing it with the following baselines.

  • Maximum mean discrepancy (MMD) (Kang et al., 2019), which is a broadly used metric for measuring discrepancies of two distributions. MMD-based adaptation method learns latent representations so that the selected out-of-domain source examples and the target examples have small MMD in the latent space.

  • Adversarial adaptation (AD) (Ganin et al., 2016), which learns latent representations so that a domain discriminator cannot tell whether an example is from the source or from the target.

Table 3 shows the results. As can be seen, our adaptation method works better than the baselines. The reason is: The domain distance metric in our method is learned using many self-supervised training examples; it can better measure domain discrepancy and facilitate domain adaptation.

4.3.3 Ablation on the Selection Mechanism

We perform an ablation study of the selection mechanism in our framework by replacing it with the following baselines.

  • Each source example is associated with a weight in [0,1]. A larger weight indicates the example is more likely to be in the target domain. We learn these data weights by minimizing the validation loss (WMVL) of the target model.

  • We use an ℋ-divergence (Elsahar and Gallé, 2019) based metric to measure domain similarity between a source example and the target dataset.

Table 3 shows the results. As can be seen, our selection mechanism works better than the baselines. The reason is: In our framework, the domain distance network (DSN) is learned in a discriminative way by performing classification on self-supervised training examples. Discriminative training can enable the DSN to better distinguish in-domain source examples from out-of-domain ones. In contrast, the selection mechanisms in the two baselines lack discriminability.

4.3.4 Ablation on the Tradeoff Parameter λ

We investigate how the performance of our method is affected by the tradeoff parameter λ. Figure 3 shows the test accuracy on ImageCLEF. As can be seen, when λ increases from 0.01 to 0.1, the accuracy improves. This is because the selected and adapted source data is used as additional training resources for the target model. However, as λ further increases, the accuracy drops. This is because the source data is not as reliable as the target data. An excessively large λ renders too much emphasis to be put on less-reliable source data.

Figure 3: 

How the micro F1 on ChemProt changes with λ.

Figure 3: 

How the micro F1 on ChemProt changes with λ.

Close modal

4.3.5 Ablation on the Train-Validate Ratio

In the next ablation study, we investigate how the performance of our method varies under different split ratios between target training and validation datasets. The study was performed on the text classification task. Table 5 shows the results. As can be seen, a more balanced ratio (e.g., 1:1) yields better performance. When the ratio is largely imbalanced (e.g., 1:9, 1:4, 1:0.1, 1:0.25), the performance is worse. If the target training dataset Dt(tr) is much smaller than the target validation dataset Dt(val), the target model’s weight parameters U, which are trained on Dt(tr), will not be sufficiently trained due to the lack of training data and thereby yield poorer performance. On the other hand, if Dt(val) is much smaller than Dt(tr), the weights A of SSL training examples, which are optimized by minimizing the loss on Dt(val), will not be sufficiently optimized due to the lack of data and thereby yield worse performance as well. Note that since Dt(val) and Dt(tr) are obtained by splitting the original target training dataset, it is always possible to obtain a balanced split.

Table 5: 

Text classification results of our method under different split ratios between target training and validation sets.

Train-val ratioChemProtRCTACL-ARCSciERCHyperPartisanAGNewsHelpfulnessIMDBAverage
1:9 79.40.4 86.90.1 73.72.7 78.51.3 87.22.3 93.00.2 65.53.0 93.30.1 82.2 
1:4 82.00.4 87.80.1 75.13.9 81.01.5 89.63.0 94.10.1 68.22.7 94.10.2 84.0 
1:2 84.20.1 89.10.1 76.45.0 83.20.9 91.74.2 94.50.1 70.00.3 95.90.2 85.6 
1:1 87.10.2 90.40.1 77.62.3 84.40.7 92.41.2 95.70.1 70.90.8 96.60.1 86.9 
1:0.5 86.80.4 90.60.1 77.12.9 84.01.2 92.81.7 95.60.1 70.41.1 96.10.1 86.7 
1:0.25 86.40.2 89.90.1 76.94.2 83.81.6 91.91.9 95.20.2 69.91.6 95.30.2 86.2 
1:0.1 85.30.5 88.50.1 76.32.8 82.60.9 91.72.2 94.90.1 69.40.8 95.00.1 85.5 
Train-val ratioChemProtRCTACL-ARCSciERCHyperPartisanAGNewsHelpfulnessIMDBAverage
1:9 79.40.4 86.90.1 73.72.7 78.51.3 87.22.3 93.00.2 65.53.0 93.30.1 82.2 
1:4 82.00.4 87.80.1 75.13.9 81.01.5 89.63.0 94.10.1 68.22.7 94.10.2 84.0 
1:2 84.20.1 89.10.1 76.45.0 83.20.9 91.74.2 94.50.1 70.00.3 95.90.2 85.6 
1:1 87.10.2 90.40.1 77.62.3 84.40.7 92.41.2 95.70.1 70.90.8 96.60.1 86.9 
1:0.5 86.80.4 90.60.1 77.12.9 84.01.2 92.81.7 95.60.1 70.41.1 96.10.1 86.7 
1:0.25 86.40.2 89.90.1 76.94.2 83.81.6 91.91.9 95.20.2 69.91.6 95.30.2 86.2 
1:0.1 85.30.5 88.50.1 76.32.8 82.60.9 91.72.2 94.90.1 69.40.8 95.00.1 85.5 

4.3.6 A Bi-level Optimization Based Ablation Setting

We also perform an ablation study which reduces the proposed four-level optimization problem to a BLO problem. The study was conducted on the text classification task. For each source example ds, i, we learn a weight bi ∈ [0,1]. A larger weight indicates that ds, i is more likely to be in the target domain. Let B={bi}i=1Ns where Ns is the number of source examples. We use the Transformer (Vaswani et al., 2017) T to perform domain adaptation. It takes a source text t as input and generates an adapted text f(t, T) which is expected to be in the target domain. The Gumbel-Softmax (Jang et al., 2017) trick is used to deal with the nondifferentiability of generated texts. At the lower level in the BLO formulation, we train the target model U. We define the training loss on selected source data as:
(18)
and the training loss on adapted source data as:
(19)
The optimization problem at this level is:
(20)
where Lt(U) is the training loss on target training data, as defined in Eq.(3). At the upper level, we evaluate U*(B, T) on the target validation set Dt(val) and update B and T by minimizing the validation loss:
(21)
The overall BLO formulation is:
(22)
Table 6 shows the results. As can be seen, this BLO method performs worse than our method. The reason is: Our method uses self-supervised learning to learn domain discrepancy (in stage I) and leverages the learned domain discrepancy metric to perform domain adaptation (in stage II). Such mechanisms are lacking in the BLO method. This further demonstrates the necessity of stage I and II in our method.
Table 6: 

Ablation study results of the BLO method on text classification.

MethodChemProtRCTACL-ARCSciERCHyperPartisanAGNewsHelpfulnessIMDBAverage
BLO 85.30.4 88.70.1 75.42.9 83.11.1 90.74.5 94.90.2 68.32.2 95.00.1 85.2 
Ours 87.10.2 90.40.1 77.62.3 84.40.7 92.41.2 95.70.1 70.90.8 96.60.1 86.9 
MethodChemProtRCTACL-ARCSciERCHyperPartisanAGNewsHelpfulnessIMDBAverage
BLO 85.30.4 88.70.1 75.42.9 83.11.1 90.74.5 94.90.2 68.32.2 95.00.1 85.2 
Ours 87.10.2 90.40.1 77.62.3 84.40.7 92.41.2 95.70.1 70.90.8 96.60.1 86.9 

4.4 Human Evaluation

We perform a human evaluation on whether the identified in-domain and out-of-domain source examples are indeed in or out of the target domain. The study is performed on ChemProt, RCT, and ACL-ARC. For each dataset, we randomly sample 200 source texts. Three undergraduates were asked to label whether these source texts are in-domain or out-of-domain. Majority vote is leveraged to decide the final label. The Kappa score among the annotations is 0.75, which indicates a strong level of agreement among the annotators. Different methods are applied to predict whether each source text is in-domain or not. Table 7 shows the results. Our method achieves the best accuracy in identifying in-domain and out-of-domain source examples, due to its mechanism of learning the domain distance network in a discriminative way (as analyzed in Section 4.3).

Table 7: 

Human evaluation results.

MethodAccuracy
BO (Ruder and Plank, 201779.9 
MMG (Wang et al., 2019a82.1 
LSI (Huan et al., 202183.9 
Separate 83.5 
WMVL 81.7 
H-Divergence 84.0 
 
Our full method 88.2 
MethodAccuracy
BO (Ruder and Plank, 201779.9 
MMG (Wang et al., 2019a82.1 
LSI (Huan et al., 202183.9 
Separate 83.5 
WMVL 81.7 
H-Divergence 84.0 
 
Our full method 88.2 

We propose a framework for simultaneous selection and adaptation of source examples in transfer learning. Our method automatically identifies which source examples are in or out of the target domain, and performs example-specific operations (either selection or adaptation). This is different from previous methods, which 1) discard out-of-domain source examples, leading to information loss; or 2) try to adapt all source examples into the target domain, incurring a risk of moving in-domain source examples outside the target domain. Our framework is based on four-level optimization. Experiments on text classification and visual question answering demonstrate the effectiveness of our proposed method.

Amittai
Axelrod
,
Xiaodong
He
, and
Jianfeng
Gao
.
2011
.
Domain adaptation via pseudo in-domain data selection
. In
Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing
, pages
355
362
.
Mathilde
Bateson
,
Hoel
Kervadec
,
Jose
Dolz
,
Herve
Lombaert
, and
Ismail Ben
Ayed
.
2020
.
Source-relaxed domain adaptation for image segmentation
.
CoRR
,
abs/2005.03697
.
Atilim Gunes
Baydin
,
Robert
Cornish
,
David
Martínez-Rubio
,
Mark
Schmidt
, and
Frank D.
Wood
.
2017
.
Online learning rate adaptation with hypergradient descent
.
CoRR
,
abs/1703.04782
.
Shai
Ben-David
,
John
Blitzer
,
Koby
Crammer
,
Alex
Kulesza
,
Fernando
Pereira
, and
Jennifer Wortman
Vaughan
.
2010
.
A theory of learning from different domains
.
Machine Learning
,
79
(
1
):
151
175
.
Karsten M.
Borgwardt
,
Arthur
Gretton
,
Malte J.
Rasch
,
Hans-Peter
Kriegel
,
Bernhard
Schölkopf
, and
Alex J.
Smola
.
2006
.
Integrating structured biological data by kernel maximum mean discrepancy
.
Bioinformatics
,
22
(
14
):
e49–e57
. ,
[PubMed]
Konstantinos
Bousmalis
,
Nathan
Silberman
,
David
Dohan
,
Dumitru
Erhan
, and
Dilip
Krishnan
.
2017
.
Unsupervised pixel-level domain adaptation with generative adversarial networks
. In
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition
, pages
3722
3731
.
Ting
Chen
,
Simon
Kornblith
,
Mohammad
Norouzi
, and
Geoffrey
Hinton
.
2020
.
A simple framework for contrastive learning of visual representations
.
arXiv preprint arXiv:2002.05709
.
Michael
Cogswell
,
Faruk
Ahmed
,
Ross
Girshick
,
Larry
Zitnick
, and
Dhruv
Batra
.
2015
.
Reducing overfitting in deep networks by decorrelating representations
.
arXiv preprint arXiv:1511.06068
.
Franck
Dernoncourt
and
Ji
Young Lee
.
2017
.
PubMed 200k RCT: A dataset for sequential sentence classification in medical abstracts
. In
Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 2: Short Papers)
, pages
308
313
,
Taipei, Taiwan
.
Asian Federation of Natural Language Processing
.
Lixin
Duan
,
Dong
Xu
, and
Ivor Wai-Hung
Tsang
.
2012
.
Domain adaptation from multiple sources: A domain-dependent regularization approach
.
IEEE Transactions on Neural Networks and Learning Systems
,
23
(
3
):
504
518
. ,
[PubMed]
Hady
Elsahar
and
Matthias
Gallé
.
2019
.
To annotate or not? Predicting performance drop under domain shift
. In
Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)
, pages
2163
2173
,
Hong Kong, China
.
Association for Computational Linguistics
.
Matthias
Feurer
,
Jost
Springenberg
, and
Frank
Hutter
.
2015
.
Initializing Bayesian hyperparameter optimization via meta-learning
. In
Proceedings of the AAAI Conference on Artificial Intelligence
, volume
29
.
Chelsea
Finn
,
Pieter
Abbeel
, and
Sergey
Levine
.
2017
.
Model-agnostic meta-learning for fast adaptation of deep networks
. In
Proceedings of the 34th International Conference on Machine Learning-Volume 70
, pages
1126
1135
.
JMLR.org
.
George
Foster
,
Cyril
Goutte
, and
Roland
Kuhn
.
2010
.
Discriminative instance weighting for domain adaptation in statistical machine translation
. In
Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing
, pages
451
459
.
Yaroslav
Ganin
,
Evgeniya
Ustinova
,
Hana
Ajakan
,
Pascal
Germain
,
Hugo
Larochelle
,
François
Laviolette
,
Mario
Marchand
, and
Victor
Lempitsky
.
2016
.
Domain-adversarial training of neural networks
.
The Journal of Machine Learning Research
,
17
(
1
):
2096
2030
.
Tianyu
Gao
,
Xingcheng
Yao
, and
Danqi
Chen
.
2021
.
SimCSE: Simple contrastive learning of sentence embeddings
. In
Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing
, pages
6894
6910
,
Online and Punta Cana, Dominican Republic
.
Association for Computational Linguistics
.
Weifeng
Ge
and
Yizhou
Yu
.
2017
.
Borrowing treasures from the wealthy: Deep transfer learning through selective joint fine-tuning
.
CoRR
,
abs/1702.08690
.
Saeed
Ghadimi
and
Mengdi
Wang
.
2018
.
Approximation methods for bilevel programming
.
arXiv preprint arXiv:1802.02246
.
Cyril
Goutte
and
Eric
Gaussier
.
2005
.
A probabilistic interpretation of precision, recall and f-score, with implication for evaluation
. In
European Conference on Information Retrieval
.
Riccardo
Grazzi
,
Luca
Franceschi
,
Massimiliano
Pontil
, and
Saverio
Salzo
.
2020
.
On the iteration complexity of hypergradient computation
. In
International Conference on Machine Learning
, pages
3748
3758
.
PMLR
.
Han
Guo
,
Ramakanth
Pasunuru
, and
Mohit
Bansal
.
2019
.
Autosem: Automatic task selection and mixing in multi-task learning
.
CoRR
,
abs/1904.04153
.
Suchin
Gururangan
,
Ana
Marasović
,
Swabha
Swayamdipta
,
Kyle
Lo
,
Iz
Beltagy
,
Doug
Downey
, and
Noah A.
Smith
.
2020
.
Don’t stop pretraining: Adapt language models to domains and tasks
. In
Proceedings of ACL
.
Kaiming
He
,
Haoqi
Fan
,
Yuxin
Wu
,
Saining
Xie
, and
Ross
Girshick
.
2019
.
Momentum contrast for unsupervised visual representation learning
.
arXiv preprint arXiv:1911.05722
.
Kaiming
He
,
Xiangyu
Zhang
,
Shaoqing
Ren
, and
Jian
Sun
.
2016
.
Deep residual learning for image recognition
. In
CVPR
.
Ruining
He
and
Julian
McAuley
.
2016
.
Ups and downs: Modeling the visual evolution of fashion trends with one-class collaborative filtering
. In
WWW
.
Xuehai
He
,
Yichen
Zhang
,
Luntian
Mou
,
Eric
Xing
, and
Pengtao
Xie
.
2020
.
Pathvqa: 30000+ questions for medical visual question answering
.
arXiv preprint arXiv:2003.10286
.
Judy
Hoffman
,
Eric
Tzeng
,
Taesung
Park
,
Jun-Yan
Zhu
,
Phillip
Isola
,
Kate
Saenko
,
Alexei
Efros
, and
Trevor
Darrell
.
2018
.
Cycada: Cycle-consistent adversarial domain adaptation
. In
International Conference on Machine Learning
, pages
1989
1998
.
PMLR
.
Zhiting
Hu
,
Bowen
Tan
,
Ruslan
Salakhutdinov
,
Tom M.
Mitchell
, and
Eric P.
Xing
.
2019
.
Learning data manipulation for augmentation and weighting
.
CoRR
,
abs/1910.12795
.
Zhaoxin
Huan
,
Yulong
Wang
,
Yong
He
,
Xiaolu
Zhang
,
Chilin
Fu
,
Weichang
Wu
,
Jun
Zhou
,
Ke
Ding
,
Liang
Zhang
, and
Linjian
Mo
.
2021
.
Learning to select instance: Simultaneous transfer learning and clustering
. In
Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval
, pages
1950
1954
.
Jiayuan
Huang
,
Arthur
Gretton
,
Karsten
Borgwardt
,
Bernhard
Schölkopf
, and
Alex
Smola
.
2006
.
Correcting sample selection bias by unlabeled data
.
Advances in Neural Information Processing Systems
,
19
:
601
608
.
Eric
Jang
,
Shixiang
Gu
, and
Ben
Poole
.
2017
.
Categorical reparametrization with gumble-softmax
. In
International Conference on Learning Representations (ICLR 2017)
.
OpenReview. net
.
Kaiyi
Ji
,
Junjie
Yang
, and
Yingbin
Liang
.
2021
.
Bilevel optimization: Convergence analysis and enhanced design
. In
International Conference on Machine Learning
, pages
4882
4892
.
PMLR
.
Jing
Jiang
and
ChengXiang
Zhai
.
2007
.
Instance weighting for domain adaptation in nlp
.
ACL
.
David
Jurgens
,
Srijan
Kumar
,
Raine
Hoover
,
Daniel A.
McFarland
, and
Dan
Jurafsky
.
2018
.
Measuring the evolution of a scientific field through citation frames
.
TACL
.
Guoliang
Kang
,
Lu
Jiang
,
Yi
Yang
, and
Alexander G.
Hauptmann
.
2019
.
Contrastive adaptation network for unsupervised domain adaptation
. In
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition
, pages
4893
4902
.
Johannes
Kiesel
,
Maria
Mestre
,
Rishabh
Shukla
,
Emmanuel
Vincent
,
Payam
Adineh
,
David
Corney
,
Benno
Stein
, and
Martin
Potthast
.
2019
.
SemEval-2019 Task 4: Hyperpartisan news detection
. In
SemEval
.
Jin-Hwa
Kim
,
Jaehyun
Jun
, and
Byoung-Tak
Zhang
.
2018
.
Bilinear attention networks
. In
NIPS
.
Taekyung
Kim
and
Changick
Kim
.
2020
.
Attract, perturb, and explore: Learning a feature alignment network for semi-supervised domain adaptation
.
Diederik P.
Kingma
and
Jimmy
Ba
.
2014
.
Adam: A method for stochastic optimization
.
arXiv preprint arXiv:1412.6980
.
Jens
Kringelum
,
Sonny Kim
Kjærulff
,
Søren
Brunak
,
Ole
Lund
,
Tudor I.
Oprea
, and
Olivier
Taboureau
.
2016
.
ChemProt-3.0: A global chemical biology diseases mapping
. In
Database
. ,
[PubMed]
Jian
Liang
,
Dapeng
Hu
, and
Jiashi
Feng
.
2021
.
Domain adaptation with auxiliary target domain-oriented classifier
. In
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition
, pages
16632
16642
.
Hanxiao
Liu
,
Karen
Simonyan
, and
Yiming
Yang
.
2018
.
Darts: Differentiable architecture search
.
arXiv preprint arXiv:1806.09055
.
Hong
Liu
,
Mingsheng
Long
,
Jianmin
Wang
, and
Michael
Jordan
.
2019a
.
Transferable adversarial training: A general approach to adapting deep classifiers
. In
Proceedings of the 36th International Conference on Machine Learning
, volume
97
, pages
4013
4022
.
PMLR
.
Miaofeng
Liu
,
Yan
Song
,
Hongbin
Zou
, and
Tong
Zhang
.
2019b
.
Reinforced training data selection for domain adaptation
. In
Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics
,
Florence, Italy
.
Association for Computational Linguistics
.
Risheng
Liu
,
Yaohua
Liu
,
Shangzhi
Zeng
, and
Jin
Zhang
.
2021
.
Towards gradient-based bilevel optimization with non-convex followers and beyond
. In
Advances in Neural Information Processing Systems
,
34
.
Yinhan
Liu
,
Myle
Ott
,
Naman
Goyal
,
Jingfei
Du
,
Mandar
Joshi
,
Danqi
Chen
,
Omer
Levy
,
Mike
Lewis
,
Luke
Zettlemoyer
, and
Veselin
Stoyanov
.
2019c
.
Roberta: A robustly optimized bert pretraining approach
.
arXiv preprint arXiv:1907.11692
.
Kyle
Lo
,
Lucy Lu
Wang
,
Mark
Neumann
,
Rodney
Kinney
, and
Dan S.
Weld
.
2019
.
S2orc: The semantic scholar open research corpus
.
arXiv preprint arXiv:1911.02782
.
Mingsheng
Long
,
Yue
Cao
,
Jianmin
Wang
, and
Michael
Jordan
.
2015
.
Learning transferable features with deep adaptation networks
. In
International Conference on Machine Learning
, pages
97
105
.
Mingsheng
Long
,
Zhangjie
Cao
,
Jianmin
Wang
, and
Michael I.
Jordan
.
2017a
.
Conditional adversarial domain adaptation
.
arXiv preprint arXiv:1705.10667
.
Mingsheng
Long
,
Jianmin
Wang
,
Guiguang
Ding
,
Jiaguang
Sun
, and
Philip
S. Yu
.
2013
.
Transfer feature learning with joint distribution adaptation
. In
Proceedings of the IEEE International Conference on Computer Vision
, pages
2200
2207
.
Mingsheng
Long
,
Han
Zhu
,
Jianmin
Wang
, and
Michael I.
Jordan
.
2017b
.
Deep transfer learning with joint adaptation networks
. In
International conference on machine learning
, pages
2208
2217
.
PMLR
.
I.
Loshchilov
and
F.
Hutter
.
2017
.
Fixing weight decay regularization in Adam
.
ArXiv
,
abs/1711.05101
.
Yi
Luan
,
Luheng
He
,
Mari
Ostendorf
, and
Hannaneh
Hajishirzi
.
2018
.
Multi-task identification of entities, relations, and coreference for scientific knowledge graph construction
. In
Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing
, pages
3219
3232
,
Brussels, Belgium
,
Association for Computational Linguistics
.
Ping
Luo
,
Fuzhen
Zhuang
,
Hui
Xiong
,
Yuhong
Xiong
, and
Qing
He
.
2008
.
Transfer learning from multiple source domains via consensus regularization
. In
Proceedings of the 17th ACM Conference on Information and Knowledge Management
, pages
103
112
.
Zelun
Luo
,
Yuliang
Zou
,
Judy
Hoffman
, and
Li
Fei-Fei
.
2017
.
Label efficient learning of transferable representations across domains and tasks
.
arXiv preprint arXiv:1712.00123
.
Andrew L.
Maas
,
Raymond E.
Daly
,
Peter T.
Pham
,
Dan
Huang
,
Andrew Y.
Ng
, and
Christopher
Potts
.
2011
.
Learning word vectors for sentiment analysis
. In
ACL
.
Mateusz
Malinowski
and
Mario
Fritz
.
2014
.
A multi-world approach to question answering about real-world scenes based on uncertain input
. In
NIPS
.
Julian
McAuley
,
Christopher
Targett
,
Qinfeng
Shi
, and
Anton
Van Den Hengel
.
2015
.
Image-based recommendations on styles and substitutes
. In
ACM SIGIR
.
Paulius
Micikevicius
,
Sharan
Narang
,
Jonah
Alben
,
Gregory
Diamos
,
Erich
Elsen
,
David
Garcia
,
Boris
Ginsburg
,
Michael
Houston
,
Oleksii
Kuchaiev
,
Ganesh
Venkatesh
, and
Hao
Wu
.
2017
.
Mixed precision training
.
arXiv preprint arXiv:1710.03740
.
Lilyana
Mihalkova
,
Tuyen
Huynh
, and
Raymond J.
Mooney
.
2007
.
Mapping and revising markov logic networks for transfer learning
. In
Proceedings of AAAI
. volume
7
, pages
608
614
.
Yu
Mitsuzumi
,
Go
Irie
,
Daiki
Ikami
, and
Takashi
Shibata
.
2021
.
Generalized domain adaptation
. In
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition
, pages
1084
1093
.
Harsh
Mohan
.
2015
.
Textbook of Pathology
.
Jaypee Brothers Medical Publishers
.
Robert C.
Moore
and
Will
Lewis
.
2010
.
Intelligent selection of language model training data
.
Saeid
Motiian
,
Quinn
Jones
,
Seyed
Iranmanesh
, and
Gianfranco
Doretto
.
2017
.
Few-shot adversarial domain adaptation
. In
Advances in Neural Information Processing Systems
,
30
.
Jiquan
Ngiam
,
Daiyi
Peng
,
Vijay
Vasudevan
,
Simon
Kornblith
,
Quoc V.
Le
, and
Ruoming
Pang
.
2018
.
Domain adaptive transfer learning with specialist models
.
arXiv preprint arXiv:1811.07056
.
Alexandru
Niculescu-Mizil
and
Rich
Caruana
.
2007
.
Inductive transfer for Bayesian network structure learning
. In
Artificial Intelligence and Statistics
, pages
339
346
.
PMLR
.
Sinno Jialin
Pan
,
Ivor W.
Tsang
,
James T.
Kwok
, and
Qiang
Yang
.
2010
.
Domain adaptation via transfer component analysis
.
IEEE Transactions on Neural Networks
,
22
(
2
):
199
210
. ,
[PubMed]
Sinno Jialin
Pan
and
Qiang
Yang
.
2009
.
A survey on transfer learning
.
IEEE Transactions on Knowledge and Data Engineering
,
22
(
10
):
1345
1359
.
Kishore
Papineni
,
Salim
Roukos
,
Todd
Ward
, and
Wei-Jing
Zhu
.
2002
.
BLEU: A method for automatic evaluation of machine translation
. In
ACL
.
Yash
Patel
,
Kashyap
Chitta
, and
Bhavan
Jasani
.
2018
.
Learning sampling policies for domain adaptation
.
CoRR
,
abs/1805.07641
.
Jeffrey
Pennington
,
Richard
Socher
, and
Christopher
Manning
.
2014
.
GloVe: Global vectors for word representation
. In
EMNLP
.
Lorien Y.
Pratt
.
1993
.
Discriminability-based transfer between neural networks
. In
Advances in Neural Information Processing Systems
, pages
204
204
.
Chen
Qu
,
Feng
Ji
,
Minghui
Qiu
,
Liu
Yang
,
Zhiyu
Min
,
Haiqing
Chen
,
Jun
Huang
, and
W.
Bruce Croft
.
2018
.
Learning to selectively transfer: Reinforced transfer learning for deep text matching
.
CoRR
,
abs/1812.11561
.
Mengye
Ren
,
Wenyuan
Zeng
,
Bin
Yang
, and
Raquel
Urtasun
.
2018
.
Learning to reweight examples for robust deep learning
.
arXiv preprint arXiv:1803.09050
.
Zhongzheng
Ren
,
Raymond
Yeh
, and
Alexander
Schwing
.
2020
.
Not all unlabeled data are equal: Learning to weight data in semi-supervised learning
. In
Advances in Neural Information Processing Systems
, volume
33
, pages
21786
21797
.
Curran Associates, Inc.
Sebastian
Ruder
and
Barbara
Plank
.
2017
.
Learning to select data for transfer learning with bayesian optimization
.
CoRR
,
abs/1707.05246
.
Kuniaki
Saito
,
Donghyun
Kim
,
Stan
Sclaroff
,
Trevor
Darrell
, and
Kate
Saenko
.
2019
.
Semi-supervised domain adaptation via minimax entropy
. In
Proceedings of the IEEE/CVF International Conference on Computer Vision
, pages
8050
8058
.
Swami
Sankaranarayanan
,
Yogesh
Balaji
,
Carlos D.
Castillo
, and
Rama
Chellappa
.
2018
.
Generate to adapt: Aligning domains using generative adversarial networks
. In
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition
, pages
8503
8512
.
Jun
Shu
,
Qi
Xie
,
Lixuan
Yi
,
Qian
Zhao
,
Sanping
Zhou
,
Zongben
Xu
, and
Deyu
Meng
.
2019
.
Meta-weight-net: Learning an explicit mapping for sample weighting
. In
Advances in Neural Information Processing Systems
, pages
1919
1930
.
Sunit
Sivasankaran
,
Emmanuel
Vincent
, and
Irina
Illina
.
2017
.
Discriminative importance weighting of augmented training data for acoustic model training
. In
2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
, pages
4885
4889
.
IEEE
.
Yan
Song
,
Prescott
Klassen
,
Fei
Xia
, and
Chunyu
Kit
.
2012
.
Entropy-based training data selection for domain adaptation
. In
Proceedings of COLING 2012: Posters
, pages
1191
1200
.
Felipe Petroski
Such
,
Aditya
Rawal
,
Joel
Lehman
,
Kenneth O.
Stanley
, and
Jeff
Clune
.
2019
.
Generative teaching networks: Accelerating neural architecture search by learning to generate synthetic training data
.
CoRR
,
abs/1912.07768
.
Baochen
Sun
and
Kate
Saenko
.
2016
.
Deep coral: Correlation alignment for deep domain adaptation
. In
European Conference on Computer Vision
, pages
443
450
.
Springer
.
Qian
Sun
,
Rita
Chattopadhyay
,
Sethuraman
Panchanathan
, and
Jieping
Ye
.
2011
.
A two-stage weighting framework for multi-source domain adaptation
.
Advances in Neural Information Processing Systems
,
24
:
505
513
.
Hao
Tan
and
Mohit
Bansal
.
2019
.
Lxmert: Learning cross-modality encoder representations from transformers
.
arXiv preprint arXiv:1908.07490
.
Hui
Tang
,
Ke
Chen
, and
Kui
Jia
.
2020
.
Unsupervised domain adaptation via structurally regularized deep clustering
.
Hui
Tang
and
Kui
Jia
.
2019
.
Discriminative adversarial domain adaptation
.
CoRR
,
abs/1911.12036
.
Tatiana
Tommasi
,
Francesco
Orabona
, and
Barbara
Caputo
.
2010
.
Safety in numbers: Learning categories from few examples with multi model knowledge transfer
. In
2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition
, pages
3081
3088
.
IEEE
.
Eric
Tzeng
,
Judy
Hoffman
,
Kate
Saenko
, and
Trevor
Darrell
.
2017
.
Adversarial discriminative domain adaptation
.
arXiv preprint arXiv:1702.05464
.
Ashish
Vaswani
,
Noam
Shazeer
,
Niki
Parmar
,
Jakob
Uszkoreit
,
Llion
Jones
,
Aidan N.
Gomez
,
Łukasz
Kaiser
, and
Illia
Polosukhin
.
2017
.
Attention is all you need
. In
Advances in Neural Information Processing Systems
, pages
5998
6008
.
B.
Wang
,
Minghui
Qiu
,
Xisen
Wang
,
Y.
Li
,
Y.
Gong
,
X.
Zeng
,
J.
Huang
,
Bo
Zheng
,
Deng
Cai
, and
Jingren
Zhou
.
2019a
.
A minimax game for instance based selective transfer learning
.
Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
.
Jindong
Wang
,
Yiqiang
Chen
,
Shuji
Hao
,
Wenjie
Feng
, and
Zhiqi
Shen
.
2017a
.
Balanced distribution adaptation for transfer learning
. In
2017 IEEE International Conference on Data Mining (ICDM)
, pages
1129
1134
.
IEEE
.
Rui
Wang
,
Masao
Utiyama
,
Lemao
Liu
,
Kehai
Chen
, and
Eiichiro
Sumita
.
2017b
.
Instance weighting for neural machine translation domain adaptation
. In
Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing
, pages
1482
1488
,
Copenhagen, Denmark
.
Association for Computational Linguistics
.
Xinyi
Wang
,
Hieu
Pham
,
Paul
Michel
,
Antonios
Anastasopoulos
,
Jaime
Carbonell
, and
Graham
Neubig
.
2020a
.
Optimizing data usage via differentiable rewards
. In
International Conference on Machine Learning
, pages
9983
9995
.
PMLR
.
Yulin
Wang
,
Jiayi
Guo
,
Shiji
Song
, and
Gao
Huang
.
2020b
.
Meta-semi: A meta-learning approach for semi-supervised learning
.
CoRR
,
abs/2007.02394
. https://arxiv.org/abs/2007.02394
Yunyun
Wang
,
Dan
Zhao
,
Yun
Li
,
Ke-Jia
Chen
, and
H.
Xue
.
2019b
.
The most related knowledge first: A progressive domain adaptation method
. In
PAKDD
.
Junjie
Yang
,
Kaiyi
Ji
, and
Yingbin
Liang
.
2021
.
Provably faster algorithms for bilevel optimization
.
Advances in Neural Information Processing Systems
,
34
.
Zichao
Yang
,
Xiaodong
He
,
Jianfeng
Gao
,
Li
Deng
, and
Alex
Smola
.
2016
.
Stacked attention networks for image question answering
. In
CVPR
.
Rowan
Zellers
,
Ari
Holtzman
,
Hannah
Rashkin
,
Yonatan
Bisk
,
Ali
Farhadi
,
Franziska
Roesner
, and
Yejin
Choi
.
2019
.
Defending against neural fake news
.
Advances in Neural Information Processing Systems
,
32
.
Xiang
Zhang
,
Junbo Jake
Zhao
, and
Yann
LeCun
.
2015
.
Character-level convolutional networks for text classification
. In
NeurIPS
.
Yabin
Zhang
,
Hui
Tang
,
Kui
Jia
, and
Mingkui
Tan
.
2019
.
Domain-symmetric networks for adversarial domain adaptation
. In
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition
, pages
5031
5040
.
Yang
Zhang
,
Philip
David
, and
Boqing
Gong
.
2017
.
Curriculum domain adaptation for semantic segmentation of urban scenes
.
CoRR
,
abs/1707.09465
.
Fuzhen
Zhuang
,
Zhiyuan
Qi
,
Keyu
Duan
,
Dongbo
Xi
,
Yongchun
Zhu
,
Hengshu
Zhu
,
Hui
Xiong
, and
Qing
He
.
2020
.
A comprehensive survey on transfer learning
.
Proceedings of the IEEE
,
109
(
1
):
43
76
.

Author notes

Equal contribution.

Action Editor: Tim Baldwin

This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. For a full description of the license, please visit https://creativecommons.org/licenses/by/4.0/legalcode.