From neurons to Newtons: Brain evolution as a machine learning problem
We have entered a golden age of artificial intelligence research, driven mainly by the advances in the artificial neural networks over the last several decades. Applications of these techniques—to machine vision, speech recognition, autonomous vehicles, natural language, and many other domains—are coming so quickly that many observers predict that the long-elusive goal of “Artificial General Intelligence” (AGI) is within our grasp. However, we still cannot build a machine capable of building a nest, stalking prey, or loading a dishwasher. I will describe how evolution may have shaped the algorithms that the brain is using to solve some of these challenging problems. Presented in the van Vreeswijk Theoretical Neuroscience Seminar series (formerly WWTNS) on 2025-05-21. Recording duration: 00:44:20.
Reimagining the neuron as a controller: A novel model for Neuroscience and AI
We build upon and expand the efficient coding and predictive information models of neurons, presenting a novel perspective that neurons not only predict but also actively influence their future inputs through their outputs. We introduce the concept of neurons as feedback controllers of their environments, a role traditionally considered computationally demanding, particularly when the dynamical system characterizing the environment is unknown. By harnessing a novel data-driven control framework, we illustrate the feasibility of biological neurons functioning as effective feedback controllers. This innovative approach enables us to coherently explain various experimental findings that previously seemed unrelated. Our research has profound implications, potentially revolutionizing the modeling of neuronal circuits and paving the way for the creation of alternative, biologically inspired artificial neural networks.
Multimodal units fuse-then-accumulate evidence across channels
We continuously detect sensory data, like sights and sounds, and use this information to guide our behaviour. However, rather than relying on single sensory channels, which are noisy and can be ambiguous alone, we merge information across our senses and leverage this combined signal. In biological networks, this process (multisensory integration) is implemented by multimodal neurons which are often thought to receive the information accumulated by unimodal areas, and to fuse this across channels; an algorithm we term accumulate-then-fuse. However, it remains an open question how well this theory generalises beyond the classical tasks used to test multimodal integration. Here, we explore this by developing novel multimodal tasks and deploying probabilistic, artificial and spiking neural network models. Using these models we demonstrate that multimodal units are not necessary for accuracy or balancing speed/accuracy in classical multimodal tasks, but are critical in a novel set of tasks in which we comodulate signals across channels. We show that these comodulation tasks require multimodal units to implement an alternative fuse-then-accumulate algorithm, which excels in naturalistic settings and is optimal for a wide class of multimodal problems. Finally, we link our findings to experimental results at multiple levels; from single neurons to behaviour. Ultimately, our work suggests that multimodal neurons may fuse-then-accumulate evidence across channels, and provides novel tasks and models for exploring this in biological systems. Presented in the van Vreeswijk Theoretical Neuroscience Seminar series (formerly WWTNS) on 2023-10-25. Recording duration: 00:30:13.
The centrality of population-level factors to network computation is demonstrated by a versatile approach for training spiking networks
Neural activity is often described in terms of population-level factors extracted from the responses of many neurons. Factors provide a lower-dimensional description with the aim of shedding light on network computations. Yet, mechanistically, computations are performed not by continuously valued factors but by interactions among neurons that spike discretely and variably. Models provide a means of bridging these levels of description. We developed a general method for training model networks of spiking neurons by leveraging factors extracted from either data or firing-rate-based networks. In addition to providing a useful model-building framework, this formalism illustrates how reliable and continuously valued factors can arise from seemingly stochastic spiking. Our framework establishes procedures for embedding this property in network models with different levels of realism. The relationship between spikes and factors in such networks provides a foundation for interpreting (and subtly redefining) commonly used quantities such as firing rates.
Analyzing artificial neural networks to understand the brain
In the first part of this talk I will present work showing that recurrent neural networks can replicate broad behavioral patterns associated with dynamic visual object recognition in humans. An analysis of these networks shows that different types of recurrence use different strategies to solve the object recognition problem. The similarities between artificial neural networks and the brain presents another opportunity, beyond using them just as models of biological processing. In the second part of this talk, I will discuss—and solicit feedback on—a proposed research plan for testing a wide range of analysis tools frequently applied to neural data on artificial neural networks. I will present the motivation for this approach as well as the form the results could take and how this would benefit neuroscience.
Bridging the gap between artificial models and cortical circuits
Artificial neural networks simplify complex biological circuits into tractable models for computational exploration and experimentation. However, the simplification of artificial models also undermines their applicability to real brain dynamics. Typical efforts to address this mismatch add complexity to increasingly unwieldy models. Here, we take a different approach; by reducing the complexity of a biological cortical culture, we aim to distil the essential factors of neuronal dynamics and plasticity. We leverage recent advances in growing neurons from human induced pluripotent stem cells (hiPSCs) to analyse ex vivo cortical cultures with only two distinct excitatory and inhibitory neuron populations. Over 6 weeks of development, we record from thousands of neurons using high-density microelectrode arrays (HD-MEAs) that allow access to individual neurons and the broader population dynamics. We compare these dynamics to two-population artificial networks of single-compartment neurons with random sparse connections and show that they produce similar dynamics. Specifically, our model captures the firing and bursting statistics of the cultures. Moreover, tightly integrating models and cultures allows us to evaluate the impact of changing architectures over weeks of development, with and without external stimuli. Broadly, the use of simplified cortical cultures enables us to use the repertoire of theoretical neuroscience techniques established over the past decades on artificial network models. Our approach of deriving neural networks from human cells also allows us, for the first time, to directly compare neural dynamics of disease and control. We found that cultures e.g. from epilepsy patients tended to have increasingly more avalanches of synchronous activity over weeks of development, in contrast to the control cultures. Next, we will test possible interventions, in silico and in vitro, in a drive for personalised approaches to medical care. This work starts bridging an important theoretical-experimental neuroscience gap for advancing our understanding of mammalian neuron dynamics.
Towards multi-system network models for cognitive neuroscience
Artificial neural networks can be useful for studying brain functions. In cognitive neuroscience, recurrent neural networks are often used to model cognitive functions. I will first offer my opinion on what is missing in the classical use of recurrent neural networks. Then I will discuss two lines of ongoing efforts in our group to move beyond the classical recurrent neural networks by studying multi-system neural networks (the talk will focus on two-system networks). These are networks that combine modules for several neural systems, such as vision, audition, prefrontal, hippocampal systems. I will showcase how multi-system networks can potentially be constrained by experimental data in fundamental ways and at scale.
Time as a continuous dimension in natural and artificial networks
Neural representations of time are central to our understanding of the world around us. I review cognitive, neurophysiological and theoretical work that converges on three simple ideas. First, the time of past events is remembered via populations of neurons with a continuum of functional time constants. Second, these time constants evenly tile the log time axis. This results in a neural Weber-Fechner scale for time which can support behavioral Weber-Fechner laws and characteristic behavioral effects in memory experiments. Third, these populations appear as dual pairs---one type of population contains cells that change firing rate monotonically over time and a second type of population that has circumscribed temporal receptive fields. These ideas can be used to build artificial neural networks that have novel properties. Of particular interest, a convolutional neural network built using these principles can generalize to arbitrary rescaling of its inputs. That is, after learning to perform a classification task on a time series presented at one speed, it successfully classifies stimuli presented slowed down or sped up. This result illustrates the point that this confluence of ideas originating in cognitive psychology and measured in the mammalian brain could have wide-reaching impacts on AI research.
Network science and network medicine: New strategies for understanding and treating the biological basis of mental ill-health
The last twenty years have witnessed extraordinarily rapid progress in basic neuroscience, including breakthrough technologies such as optogenetics, and the collection of unprecedented amounts of neuroimaging, genetic and other data relevant to neuroscience and mental health. However, the translation of this progress into improved understanding of brain function and dysfunction has been comparatively slow. As a result, the development of therapeutics for mental health has stagnated too. One central challenge has been to extract meaning from these large, complex, multivariate datasets, which requires a shift towards systems-level mathematical and computational approaches. A second challenge has been reconciling different scales of investigation, from genes and molecules to cells, circuits, tissue, whole-brain, and ultimately behaviour. In this talk I will describe several strands of work using mathematical, statistical, and bioinformatic methods to bridge these gaps. Topics will include: using artificial neural networks to link the organization of large-scale brain connectivity to cognitive function; using multivariate statistical methods to link disease-related changes in brain networks to the underlying biological processes; and using network-based approaches to move from genetic insights towards drug discovey. Finally, I will discuss how simple organisms such as C. elegans can serve to inspire, test, and validate new methods and insights in networks neuroscience.
Analogical Reasoning with Neuro-Symbolic AI
Knowledge discovery with computers requires a huge amount of search. Analogical reasoning is effective for efficient knowledge discovery. Therefore, we proposed analogical reasoning systems based on first-order predicate logic using Neuro-Symbolic AI. Neuro-Symbolic AI is a combination of Symbolic AI and artificial neural networks and has features that are easy for human interpretation and robust against data ambiguity and errors. We have implemented analogical reasoning systems by Neuro-symbolic AI models with word embedding which can represent similarity between words. Using the proposed systems, we efficiently extracted unknown rules from knowledge bases described in Prolog. The proposed method is the first case of analogical reasoning based on the first-order predicate logic using deep learning.
What does the primary visual cortex tell us about object recognition?
Object recognition relies on the complex visual representations in cortical areas at the top of the ventral stream hierarchy. While these are thought to be derived from low-level stages of visual processing, this has not been shown, yet. Here, I describe the results of two projects exploring the contributions of primary visual cortex (V1) processing to object recognition using artificial neural networks (ANNs). First, we developed hundreds of ANN-based V1 models and evaluated how their single neurons approximate those in the macaque V1. We found that, for some models, single neurons in intermediate layers are similar to their biological counterparts, and that the distributions of their response properties approximately match those in V1. Furthermore, we observed that models that better matched macaque V1 were also more aligned with human behavior, suggesting that object recognition is derived from low-level. Motivated by these results, we then studied how an ANN’s robustness to image perturbations relates to its ability to predict V1 responses. Despite their high performance in object recognition tasks, ANNs can be fooled by imperceptibly small, explicitly crafted perturbations. We observed that ANNs that better predicted V1 neuronal activity were also more robust to adversarial attacks. Inspired by this, we developed VOneNets, a new class of hybrid ANN vision models. Each VOneNet contains a fixed neural network front-end that simulates primate V1 followed by a neural network back-end adapted from current computer vision models. After training, VOneNets were substantially more robust, outperforming state-of-the-art methods on a set of perturbations. While current neural network architectures are arguably brain-inspired, these results demonstrate that more precisely mimicking just one stage of the primate visual system leads to new gains in computer vision applications and results in better models of the primate ventral stream and object recognition behavior.
On the implicit bias of SGD in deep learning
Tali's work emphasized the tradeoff between compression and information preservation. In this talk I will explore this theme in the context of deep learning. Artificial neural networks have recently revolutionized the field of machine learning. However, we still do not have sufficient theoretical understanding of how such models can be successfully learned. Two specific questions in this context are: how can neural nets be learned despite the non-convexity of the learning problem, and how can they generalize well despite often having more parameters than training data. I will describe our recent work showing that gradient-descent optimization indeed leads to 'simpler' models, where simplicity is captured by lower weight norm and in some cases clustering of weight vectors. We demonstrate this for several teacher and student architectures, including learning linear teachers with ReLU networks, learning boolean functions and learning convolutional pattern detection architectures.
Towards a neurally mechanistic understanding of visual cognition
I am interested in developing a neurally mechanistic understanding of how primate brains represent the world through its visual system and how such representations enable a remarkable set of intelligent behaviors. In this talk, I will primarily highlight aspects of my current research that focuses on dissecting the brain circuits that support core object recognition behavior (primates’ ability to categorize objects within hundreds of milliseconds) in non-human primates. On the one hand, my work empirically examines how well computational models of the primate ventral visual pathways embed knowledge of the visual brain function (e.g., Bashivan*, Kar*, DiCarlo, Science, 2019). On the other hand, my work has led to various functional and architectural insights that help improve such brain models. For instance, we have exposed the necessity of recurrent computations in primate core object recognition (Kar et al., Nature Neuroscience, 2019), one that is strikingly missing from most feedforward artificial neural network models. Specifically, we have observed that the primate ventral stream requires fast recurrent processing via ventrolateral PFC for robust core object recognition (Kar and DiCarlo, Neuron, 2021). In addition, I have been currently developing various chemogenetic strategies to causally target specific bidirectional neural circuits in the macaque brain during multiple object recognition tasks to further probe their relevance during this behavior. I plan to transform these data and insights into tangible progress in neuroscience via my collaboration with various computational groups and building improved brain models of object recognition. I hope to end the talk with a brief glimpse of some of my planned future work!
Artificial neural networks do not adequately mimic whatever is going on in the real brain
One may think that Deep Learning technology works in ways that are similar to the human brain. This is not really true. Our best AI technology still does not mimic the brain sufficiently well to be a match in intelligence. I will describe seven differences on how our minds work in ways diametrically opposite to those of Deep Learning technology.
Do deep learning latent spaces resemble human brain representations?
In recent years, artificial neural networks have demonstrated human-like or super-human performance in many tasks including image or speech recognition, natural language processing (NLP), playing Go, chess, poker and video-games. One remarkable feature of the resulting models is that they can develop very intuitive latent representations of their inputs. In these latent spaces, simple linear operations tend to give meaningful results, as in the well-known analogy QUEEN-WOMAN+MAN=KING. We postulate that human brain representations share essential properties with these deep learning latent spaces. To verify this, we test whether artificial latent spaces can serve as a good model for decoding brain activity. We report improvements over state-of-the-art performance for reconstructing seen and imagined face images from fMRI brain activation patterns, using the latent space of a GAN (Generative Adversarial Network) model coupled with a Variational AutoEncoder (VAE). With another GAN model (BigBiGAN), we can decode and reconstruct natural scenes of any category from the corresponding brain activity. Our results suggest that deep learning can produce high-level representations approaching those found in the human brain. Finally, I will discuss whether these deep learning latent spaces could be relevant to the study of consciousness.
A function approximation perspective on neural representations
Activity patterns of neural populations in natural and artificial neural networks constitute representations of data. The nature of these representations and how they are learned are key questions in neuroscience and deep learning. In his talk, I will describe my group's efforts in building a theory of representations as feature maps leading to sample efficient function approximation. Kernel methods are at the heart of these developments. I will present applications to deep learning and neuronal data.
A computational explanation for domain specificity in the human brain
Many regions of the human brain conduct highly specific functions, such as recognizing faces, understanding language, and thinking about other people’s thoughts. Why might this domain specific organization be a good design strategy for brains, and what is the origin of domain specificity in the first place? In this talk, I will present recent work testing whether the segregation of face and object perception in human brains emerges naturally from an optimization for both tasks. We trained artificial neural networks on face and object recognition, and found that networks were able to perform both tasks well by spontaneously segregating them into distinct pathways. Critically, networks neither had prior knowledge nor any inductive bias about the tasks. Furthermore, networks optimized on tasks which apparently do not develop specialization in the human brain, such as food or cars, and object categorization showed less task segregation. These results suggest that functional segregation can spontaneously emerge without a task-specific bias, and that the domain-specific organization of the cortex may reflect a computational optimization for the real-world tasks humans solve.
Geometry of Neural Computation Unifies Working Memory and Planning
Cognitive tasks typically require the integration of working memory, contextual processing, and planning to be carried out in close coordination. However, these computations are typically studied within neuroscience as independent modular processes in the brain. In this talk I will present an alternative view, that neural representations of mappings between expected stimuli and contingent goal actions can unify working memory and planning computations. We term these stored maps contingency representations. We developed a "conditional delayed logic" task capable of disambiguating the types of representations used during performance of delay tasks. Human behaviour in this task is consistent with the contingency representation, and not with traditional sensory models of working memory. In task-optimized artificial recurrent neural network models, we investigated the representational geometry and dynamical circuit mechanisms supporting contingency-based computation, and show how contingency representation explains salient observations of neuronal tuning properties in prefrontal cortex. Finally, our theory generates novel and falsifiable predictions for single-unit and population neural recordings.