Toward an Integration of Deep Learning and Neuroscience

By Adam Henry Marblestone, Greg Wayne, Konrad Paul Kording. 2016

683 citations recorded by the source corpus.

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) the 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. In support of these hypotheses, we argue that a range of implementations of credit assignment through multiple layers of neurons are compatible with our current knowledge of neural circuitry, and that the brain’s specialized systems can be interpreted as enabling efficient optimization for specific problem classes. 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.

Subfields: Cognitive Neuroscience, Electrical and Electronic Engineering, Representation & Self-Supervised Learning, Neuroscience

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

In Electrical and Electronic Engineering

In Representation & Self-Supervised Learning

In Neuroscience

Citations in the corpus

An overview of deep learning in medical imaging focusing on MRI (2018, background): “However, there is some interesting recent work connecting neuroscience and artificial neural networks, indicating an increase in the cross-fertilization between the two fields [43, 44, 45].”
The hardware lottery (2020, background): “Our own intelligence relies on decentralized local updates which surface a global signal in ways that are still not well understood (Lillicrap & Santoro, 2019; Marblestone et al., 2016; Bi & Poo, 1998).”
SuperSpike: Supervised Learning in Multilayer Spiking Neural Networks (2017, background): “…brain learns, several studies have begun to draw interesting parallels between the internal representations formed by deep neural networks and the recorded activity from different brain regions (Yamins et al., 2014; McClure and Kriegeskorte, 2016; McIntosh et al., 2016; Marblestone et al., 2016).”
If deep learning is the answer, what is the question? (2020, background): “However, it has yet to provide a concrete road map for systems neuroscience research 23,32–34 .”
Computational Principles of Supervised Learning in the Cerebellum (2018, neutral): “Supervised learning is also thought to play a key role in the development and maintenance of many brain functions (Doya 2000, Knudsen 1994), although the organizational principles governing the underlying neural computations have been elusive (Marblestone et al. 2016).”
Surfing Uncertainty: Prediction, Action, and the Embodied Mind. (2017, neutral): “Finally, current work suggests that brains might actually optimize many diverse cost functions that would not be reducible to prediction error minimization without explanatory loss (cf. Marblestone et al. 2016).”
Epigenetics Analysis and Integrated Analysis of Multiomics Data, Including Epigenetic Data, Using Artificial Intelligence in the Era of Precision Medicine (2019, background): “Specifically, neural networks tend to be static and symbolic, while the biological brain of most living organisms is dynamic (plastic) and analog [101].”
Deep Learning for Spatio-Temporal Data Mining: A Survey (2019, neutral): “Application domains Related works Transportation [177], [187], [116], [31], [29], [6], [32], [127], [53], [112], [26], [105], [166], [82], [23], [66], [108], [126], [81], [139], [78], [128], [146], [148], [72], [98], [189], [22], [24], [99], [96], [16], [97], [85], [15], [164], [165], [95], [140], [”
Neuroscience-Inspired Artificial Intelligence. (2017, background): “The brain does not learn by implementing a single, global optimization principle within a uniform and undifferentiated neural network (Marblestone et al., 2016).”
Switching between internal and external modes: A multiscale learning principle (2017, neutral): “…a common scalar error (such as a reward signal) that is generated in a separate neural system, but this form of learning does not solve the credit assignment problem in multilayer systems and is inefﬁcient for high-dimensional representations (Marblestone et al., 2016; Werfel, Xie, & Seung, 2005).”
Training Spiking Neural Networks Using Lessons From Deep Learning (2021, background): “than a global signal that governs a unified goal for the brain [124].”