Symmetry-Based Representations for Artificial and Biological General Intelligence.

Biological intelligence is remarkable in its ability to produce complex behavior in many diverse situations through data efficient, generalizable, and transferable skill acquisition. It is believed that learning "good" sensory representations is important for enabling this, however there is little agreement as to what a good representation should look like. In this review article we are going to argue that symmetry transformations are a fundamental principle that can guide our search for what makes a good representation. The idea that there exist transformations (symmetries) that affect some aspects of the system but not others, and their relationship to conserved quantities has become central in modern physics, resulting in a more unified theoretical framework and even ability to predict the existence of new particles. Recently, symmetries have started to gain prominence in machine learning too, resulting in more data efficient and generalizable algorithms that can mimic some of the complex behaviors produced by biological intelligence. Finally, first demonstrations of the importance of symmetry transformations for representation learning in the brain are starting to arise in neuroscience. Taken together, the overwhelming positive effect that symmetries bring to these disciplines suggest that they may be an important general framework that determines the structure of the universe, constrains the nature of natural tasks and consequently shapes both biological and artificial intelligence.

Unsupervised deep learning identifies semantic disentanglement in single inferotemporal face patch neurons.

In order to better understand how the brain perceives faces, it is important to know what objective drives learning in the ventral visual stream. To answer this question, we model neural responses to faces in the macaque inferotemporal (IT) cortex with a deep self-supervised generative model, ß-VAE, which disentangles sensory data into interpretable latent factors, such as gender or age. Our results demonstrate a strong correspondence between the generative factors discovered by ß-VAE and those coded by single IT neurons, beyond that found for the baselines, including the handcrafted state-of-the-art model of face perception, the Active Appearance Model, and deep classifiers. Moreover, ß-VAE is able to reconstruct novel face images using signals from just a handful of cells. Together our results imply that optimising the disentangling objective leads to representations that closely resemble those in the IT at the single unit level. This points at disentangling as a plausible learning objective for the visual brain.

A Computational Account of the Role of Cochlear Nucleus and Inferior Colliculus in Stabilizing Auditory Nerve Firing for Auditory Category Learning.

It is well known that auditory nerve (AN) fibers overcome bandwidth limitations through the volley principle, a form of multiplexing. What is less well known is that the volley principle introduces a degree of unpredictability into AN neural firing patterns that may be affecting even simple stimulus categorization learning. We use a physiologically grounded, unsupervised spiking neural network model of the auditory brain with spike time dependent plasticity learning to demonstrate that plastic auditory cortex is unable to learn even simple auditory object categories when exposed to the raw AN firing input without subcortical preprocessing. We then demonstrate the importance of nonplastic subcortical preprocessing within the cochlear nucleus and the inferior colliculus for stabilizing and denoising AN responses. Such preprocessing enables the plastic auditory cortex to learn efficient robust representations of the auditory object categories. The biological realism of our model makes it suitable for generating neurophysiologically testable hypotheses.

Unsupervised learning of temporal features for word categorization in a spiking neural network model of the auditory brain.

The nature of the code used in the auditory cortex to represent complex auditory stimuli, such as naturally spoken words, remains a matter of debate. Here we argue that such representations are encoded by stable spatio-temporal patterns of firing within cell assemblies known as polychronous groups, or PGs. We develop a physiologically grounded, unsupervised spiking neural network model of the auditory brain with local, biologically realistic, spike-time dependent plasticity (STDP) learning, and show that the plastic cortical layers of the network develop PGs which convey substantially more information about the speaker independent identity of two naturally spoken word stimuli than does rate encoding that ignores the precise spike timings. We furthermore demonstrate that such informative PGs can only develop if the input spatio-temporal spike patterns to the plastic cortical areas of the model are relatively stable.

Harmonic Training and the Formation of Pitch Representation in a Neural Network Model of the Auditory Brain.

Attempting to explain the perceptual qualities of pitch has proven to be, and remains, a difficult problem. The wide range of sounds which elicit pitch and a lack of agreement across neurophysiological studies on how pitch is encoded by the brain have made this attempt more difficult. In describing the potential neural mechanisms by which pitch may be processed, a number of neural networks have been proposed and implemented. However, no unsupervised neural networks with biologically accurate cochlear inputs have yet been demonstrated. This paper proposes a simple system in which pitch representing neurons are produced in a biologically plausible setting. Purely unsupervised regimes of neural network learning are implemented and these prove to be sufficient in identifying the pitch of sounds with a variety of spectral profiles, including sounds with missing fundamental frequencies and iterated rippled noises.

Learning view invariant recognition with partially occluded objects.

This paper investigates how a neural network model of the ventral visual pathway, VisNet, can form separate view invariant representations of a number of objects seen rotating together. In particular, in the current work one of the rotating objects is always partially occluded by the other objects present during training. A key challenge for the model is to link together the separate partial views of the occluded object into a single view invariant representation of that object. We show how this can be achieved by Continuous Transformation (CT) learning, which relies on spatial similarity between successive views of each object. After training, the network had developed cells in the output layer which had learned to respond invariantly to particular objects over most or all views, with each cell responding to only one object. All objects, including the partially occluded object, were individually represented by a unique subset of output cells.