Towards an integration of deep learning and neuroscience

Abstract

Neuroscience has focused on the detailed implementation of computation, studying neural codes, dynamics and circuits. In machine learning, however, artificial neural networks tend to eschew precisely designed codes, dynamics or circuits in favor of brute force optimization of a cost function, often using simple and relatively uniform initial architectures. Two recent developments have emerged within machine learning that create an opportunity to connect these seemingly divergent perspectives. First, structured architectures are used, including dedicated systems for attention, recursion and various forms of short-and long-term memory storage. Second, cost functions and training procedures have become more complex and are varied across layers and over time. Here we think about the brain in terms of these ideas. We hypothesize that (1) the brain optimizes cost functions, (2) these cost functions are diverse and differ across brain locations and over development, and (3) optimization operates within a pre-structured architecture matched to the computational problems posed by behavior. Such a heterogeneously optimized system, enabled by a series of interacting cost functions, serves to make learning data-efficient and precisely targeted to the needs of the organism. We suggest directions by which neuroscience could seek to refine and test these hypotheses.

Multi-scale brain networks

“…On the other hand, the function of network nodes and circuits as well as their biophysical attributes likely depend critically upon the scale at which a network is constructed and analyzed. Accordingly, we might also expect networks to be optimized to perform scale-specific functions (Marblestone et al, 2016), and studying a particular scale gives us a unique insight into the network architecture underpinning those functions. Ultimately, network neuroscience will need both approaches—an understanding of network function and organization at specific scales, as well as a map that bridges multiple different spatial, temporal, and topological scales.…”

Reward-based training of recurrent neural networks for cognitive and value-based tasks

“…Anatomically, the policy networks may correspond to circuits in dorsolateral prefrontal cortex, while the value networks may correspond to circuits in the orbitofrontal cortex (Schultz et al, 2000;Takahashi et al, 2011) or basal ganglia (Hikosaka et al, 2014). This architecture also provides a useful example of the hypothesis that various areas of the brain effectively optimize different cost functions (Marblestone et al, 2016): in this case, the policy network maximizes reward, while the value network minimizes the prediction error for future reward.…”

The Cluster Variation Method: A Primer for Neuroscientists

“…For this related role, the most common approach thus far has been to construct various categorization engines, interpreting real-time neurally-generated signals [35]. Most significantly, recent advances in deep learning [36] build on earlier neural network architectures [37,38,39] and use brain architecture-based principles to solve very computationally tough problems.…”