Canonical neural networks perform active inference
The free-energy principle and active inference have received a significant attention in the fields of neuroscience and machine learning. However, it remains to be established whether active inference is an apt explanation for any given neural network that actively exchanges with its environment. To address this issue, we show that a class of canonical neural networks of rate coding models implicitly performs variational Bayesian inference under a well-known form of partially observed Markov decision process model (Isomura, Shimazaki, Friston, Commun Biol, 2022). Based on the proposed theory, we demonstrate that canonical neural networks—featuring delayed modulation of Hebbian plasticity—can perform planning and adaptive behavioural control in the Bayes optimal manner, through postdiction of their previous decisions. This scheme enables us to estimate implicit priors under which the agent’s neural network operates and identify a specific form of the generative model. The proposed equivalence is crucial for rendering brain activity explainable to better understand basic neuropsychology and psychiatric disorders. Moreover, this notion can dramatically reduce the complexity of designing self-learning neuromorphic hardware to perform various types of tasks.
A generative network model of neurodevelopment
The emergence of large-scale brain networks, and their continual refinement, represent crucial developmental processes that can drive individual differences in cognition and which are associated with multiple neurodevelopmental conditions. But how does this organization arise, and what mechanisms govern the diversity of these developmental processes? There are many existing descriptive theories, but to date none are computationally formalized. We provide a mathematical framework that specifies the growth of a brain network over developmental time. Within this framework macroscopic brain organization, complete with spatial embedding of its organization, is an emergent property of a generative wiring equation that optimizes its connectivity by renegotiating its biological costs and topological values continuously over development. The rules that govern these iterative wiring properties are controlled by a set of tightly framed parameters, with subtle differences in these parameters steering network growth towards different neurodiverse outcomes. Regional expression of genes associated with the developmental simulations converge on biological processes and cellular components predominantly involved in synaptic signaling, neuronal projection, catabolic intracellular processes and protein transport. Together, this provides a unifying computational framework for conceptualizing the mechanisms and diversity of childhood brain development, capable of integrating different levels of analysis – from genes to cognition. (Pre-print: https://www.biorxiv.org/content/10.1101/2020.08.13.249391v1)
Machine Learning as a tool for positive impact : case studies from climate change
Climate change is one of our generation's greatest challenges, with increasingly severe consequences on global ecosystems and populations. Machine Learning has the potential to address many important challenges in climate change, from both mitigation (reducing its extent) and adaptation (preparing for unavoidable consequences) aspects. To present the extent of these opportunities, I will describe some of the projects that I am involved in, spanning from generative model to computer vision and natural language processing. There are many opportunities for fundamental innovation in this field, advancing the state-of-the-art in Machine Learning while ensuring that this fundamental progress translates into positive real-world impact.
Robotic mapping and generative modelling of cytokine response
We have developed a robotic platform allowing us to monitor cytokines dynamics (including IL-2, IFN-g, TNF, IL-6) of immune cells in vitro, with unprecedented resolution. To understand the complex emerging dynamics, we use interpretable machine learning techniques to build a generative model of cytokine response. We discover that, surprisingly, immune activity is encoded into one global parameter, encoding ligand antigenic properties and to a less extent ligand quantity. Based on this we build a simple interpretable model which can fully explain the broad variability of cytokines dynamics. We validate our approach using different lines of cells and different ligands. Two processes are identified, connected to timing and intensity of cytokine response, which we successfully modulate using drugs or by changing conditions such as initial T cell numbers. Our work reveals a simple "cytokine code", which can be used to better understand immune response in different contexts including immunotherapy. More generally, it reveals how robotic platforms and machine learning can be leveraged to build and validate systems biology models.