One of the most critical challenges in systems neuroscience is determining the neural code. A principled framework for addressing this can be found in information theory. With this approach, one can determine whether a proposed code can account for the stimulus-response relationship. Specifically, one can compare the transmitted information between the stimulus and the hypothesized neural code with the transmitted information between the stimulus and the behavioral response. If the former is smaller than the latter (i.e., if the code cannot account for the behavior), the code can be ruled out. The information-theoretic index most widely used in this context is Shannon's mutual information. The Shannon test, however, is not ideal for this purpose: while the codes it will rule out are truly nonviable, there will be some nonviable codes that it will fail to rule out. Here we describe a wide range of alternative indices that can be used for ruling codes out. The range includes a continuum from Shannon information to measures of the performance of a Bayesian decoder. We analyze the relationship of these indices to each other and their complementary strengths and weaknesses for addressing this problem.

Cortical neurons are predominantly excitatory and highly interconnected. In spite of this, the cortex is remarkably stable: normal brains do not exhibit the kind of runaway excitation one might expect of such a system. How does the cortex maintain stability in the face of this massive excitatory feedback? More importantly, how does it do so during computations, which necessarily involve elevated firing rates? Here we address these questions in the context of attractor networks—networks that exhibit multiple stable states, or memories. We find that such networks can be stabilized at the relatively low firing rates observed in vivo if two conditions are met: (1) the background state, where all neurons are firing at low rates, is inhibition dominated, and (2) the fraction of neurons involved in a memory is above some threshold, so that there is sufficient coupling between the memory neurons and the background. This allows “dynamical stabilization” of the attractors, meaning feedback from the pool of background neurons stabilizes what would otherwise be an unstable state. We suggest that dynamical stabilization may be a strategy used for a broad range of computations, not just those involving attractors.