A Game Theoretical Framework for Quantifying Causes in Neural Networks
Which nodes in a brain network causally influence one another, and how do such interactions utilize the underlying structural connectivity? One of the fundamental goals of neuroscience is to pinpoint such causal relations. Conventionally, these relationships are established by manipulating a node while tracking changes in another node. A causal role is then assigned to the first node if this intervention led to a significant change in the state of the tracked node. In this presentation, I use a series of intuitive thought experiments to demonstrate the methodological shortcomings of the current ‘causation via manipulation’ framework. Namely, a node might causally influence another node, but how much and through which mechanistic interactions? Therefore, establishing a causal relationship, however reliable, does not provide the proper causal understanding of the system, because there often exists a wide range of causal influences that require to be adequately decomposed. To do so, I introduce a game-theoretical framework called Multi-perturbation Shapley value Analysis (MSA). Then, I present our work in which we employed MSA on an Echo State Network (ESN), quantified how much its nodes were influencing each other, and compared these measures with the underlying synaptic strength. We found that: 1. Even though the network itself was sparse, every node could causally influence other nodes. In this case, a mere elucidation of causal relationships did not provide any useful information. 2. Additionally, the full knowledge of the structural connectome did not provide a complete causal picture of the system either, since nodes frequently influenced each other indirectly, that is, via other intermediate nodes. Our results show that just elucidating causal contributions in complex networks such as the brain is not sufficient to draw mechanistic conclusions. Moreover, quantifying causal interactions requires a systematic and extensive manipulation framework. The framework put forward here benefits from employing neural network models, and in turn, provides explainability for them.
NMC4 Short Talk: Neural Representation: Bridging Neuroscience and Philosophy
We understand the brain in representational terms. E.g., we understand spatial navigation by appealing to the spatial properties that hippocampal cells represent, and the operations hippocampal circuits perform on those representations (Moser et al., 2008). Philosophers have been concerned with the nature of representation, and recently neuroscientists entered the debate, focusing specifically on neural representations. (Baker & Lansdell, n.d.; Egan, 2019; Piccinini & Shagrir, 2014; Poldrack, 2020; Shagrir, 2001). We want to know what representations are, how to discover them in the brain, and why they matter so much for our understanding of the brain. Those questions are framed in a traditional philosophical way: we start with explanations that use representational notions, and to more deeply understand those explanations we ask, what are representations — what is the definition of representation? What is it for some bit of neural activity to be a representation? I argue that there is an alternative, and much more fruitful, approach. Rather than asking what representations are, we should ask what the use of representational *notions* allows us to do in neuroscience — what thinking in representational terms helps scientists do or explain. I argue that this framing offers more fruitful ground for interdisciplinary collaboration by distinguishing the philosophical concerns that have a place in neuroscience from those that don’t (namely the definitional or metaphysical questions about representation). And I argue for a particular view of representational notions: they allow us to impose the structure of one domain onto another as a model of its causal structue. So, e.g., thinking about the hippocampus as representing spatial properties is a way of taking structures in those spatial properties, and projecting those structures (and algorithms that would implement them) them onto the brain as models of its causal structure.
Understanding "why": The role of causality in cognition
Humans have a remarkable ability to figure out what happened and why. In this talk, I will shed light on this ability from multiple angles. I will present a computational framework for modeling causal explanations in terms of counterfactual simulations, and several lines of experiments testing this framework in the domain of intuitive physics. The model predicts people's causal judgments about a variety of physical scenes, including dynamic collision events, complex situations that involve multiple causes, omissions as causes, and causal responsibility for a system's stability. It also captures the cognitive processes underlying these judgments as revealed by spontaneous eye-movements. More recently, we have applied our computational framework to explain multisensory integration. I will show how people's inferences about what happened are well-accounted for by a model that integrates visual and auditory evidence through approximate physical simulations.
Multilevel Causal Modeling
Complex systems can be modeled at various levels of granularity, e.g., we can model a person at the cognitive level, on the neuronal level, or down to the biochemical level. When multiple models represent the same system at different scales, we would like to be able to reason about the causal effects of interventions on each level in such a way that the models remain consistent across levels. In the first part of this talk, I consider which conditions must be fulfilled for two structural equation models (SEMs) to stand in such a causally consistent relation. In the second part of the talk, I present recent work on learning causally consistent SEMs across multiple levels, distinguishing between bottom-up (micro- to macro-level) and top-down (macro- to micro-level) approaches.