[Paper Review] Explanatory models in neuroscience: Part 1 -- taking mechanistic abstraction seriously
The paper argues that certain neural network models can be mechanistic explanations of brain function under a strengthened 3M++ framework, introducing runnable abstraction and similarity transforms to map models to brain mechanisms.
Despite the recent success of neural network models in mimicking animal performance on visual perceptual tasks, critics worry that these models fail to illuminate brain function. We take it that a central approach to explanation in systems neuroscience is that of mechanistic modeling, where understanding the system is taken to require fleshing out the parts, organization, and activities of a system, and how those give rise to behaviors of interest. However, it remains somewhat controversial what it means for a model to describe a mechanism, and whether neural network models qualify as explanatory. We argue that certain kinds of neural network models are actually good examples of mechanistic models, when the right notion of mechanistic mapping is deployed. Building on existing work on model-to-mechanism mapping (3M), we describe criteria delineating such a notion, which we call 3M++. These criteria require us, first, to identify a level of description that is both abstract but detailed enough to be "runnable", and then, to construct model-to-brain mappings using the same principles as those employed for brain-to-brain mapping across individuals. Perhaps surprisingly, the abstractions required are those already in use in experimental neuroscience, and are of the kind deployed in the construction of more familiar computational models, just as the principles of inter-brain mappings are very much in the spirit of those already employed in the collection and analysis of data across animals. In a companion paper, we address the relationship between optimization and intelligibility, in the context of functional evolutionary explanations. Taken together, mechanistic interpretations of computational models and the dependencies between form and function illuminated by optimization processes can help us to understand why brain systems are built they way they are.
Motivation & Objective
- Clarify what counts as a mechanistic explanation in systems neuroscience.
- Propose an expanded Model-Mechanism-Mapping criterion (3M++) that accommodates neural networks.
- Explain how abstraction, runnable models, and similarity transforms support model-to-brain mappings.
- Provide foundational abstractions from neuroscience to ground 3M++.
Proposed method
- Review and synthesize mechanistic explanation literature (3M) and interventionist causality concepts.
- Define 3M++ by adding Predictively Adequate Runnable Abstraction (PARA) and Transform Similarity to 3M.
- Formalize how abstractions can be runnable and causally sufficient for the target phenomenon.
- Use ventral visual pathway as a case study for mapping hierarchical convolutional models to brain areas.
- Discuss how abstract neural networks relate to biological spiking networks and LN units.
Experimental results
Research questions
- RQ1What counts as explanatory when modeling brain function with computational models?
- RQ2How can neural networks be mapped to brain mechanisms while respecting abstraction levels?
- RQ3What criteria (3M++) best ensure a model is mechanistic and explanatory for neuroscience?
- RQ4When do abstractions become runnable and predictive for novel inputs?
- RQ5How do similarity transforms facilitate cross-individual brain-to-model mappings?
Key findings
- Neural networks can be explanatory models when evaluated with an expanded 3M++ framework.
- Mechanistic mapping can be achieved by runnable abstractions and similarity mappings across individuals.
- Foundational abstractions from neuroscience (e.g., LN units, retinotopy, hierarchical processing) align with computational model structures like HCNNs.
- DNNs can provide abstract mechanistic explanations in some cases, bridging model performance and brain mechanisms.
- A framework grounded in experimental neuroscience abstractions supports evaluating both old and new models for mechanistic explanation.
Better researchstarts right now
From reading papers to final review, dramatically reduce your research time.
No credit card · Free plan available
This review was created by AI and reviewed by human editors.