Meeting Summary: We aim to begin to develop theoretical foundations for explaining the kind of high-level information processing in the brain that underpins cognition in terms of the underlying physical processes. For example, we seek to understand the flexible computational architectures required to integrate multiple sensory streams whose quality varies in time with internal and social information in order to make decisions and select goals. Likewise, we seek the mathematics that captures the information processing that happens during contemplation, or when we read a book, or when we think about those in our social network.
Physically, all of these processes are fundamentally out-of-equilibrium, and dynamically changing (not at a stationary state). Cognitively, they involve dynamic changes in goals, internal states, and sensory streams, all of which require ongoing flexible changes to the underlying physical architecture of computation. Classical neuroscience theory, focusing on equilibrium information paradigms and stationary states, uses tools like efficient coding theory, which are insufficient to analyze these kinds of processes.
In this working group, we will exploit new developments in non-equilibrium statistical physics and information processing theory to develop foundations for theories of the physical processes underlying high-level computation in the brain. All computational architectures and algorithms are inherently shaped by the constraints to which their components are subject. Thus, we will focus on energetic constraints, which have been argued to play a significant role in limiting neural computation. The powerful new field of stochastic thermodynamics will play a central role. In addition, we will explore the application to the brain of new paradigms for asynchronous, distributed, computation by non-equilibrium systems such as liquid-state machines, random Boolean functions, Petri nets, concurrent programming, asynchronous control, and reservoir computation.