Brains have evolved diverse neurons with varying morphologies and dynamics that impact temporal information processing. In contrast, most neural network models use homogeneous units that vary only in spatial parameters (weights and biases). To explore the importance of temporal parameters, we trained spiking neural networks on tasks with varying temporal complexity, holding different parameter subsets constant. We found that adapting conduction delays is crucial for solving all test conditions under tight resource constraints. Remarkably, these tasks can be solved using only temporal parameters (delays and time constants) with constant weights. In more complex spatio-temporal tasks, an adaptable bursting parameter was essential. Overall, allowing adaptation of both temporal and spatial parameters enhances network robustness to noise, a vital feature for biological brains and neuromorphic computing systems. Our findings suggest that rich and adaptable dynamics may be the key for solving temporally structured tasks efficiently in evolving organisms, which would help explain the diverse physiological properties of biological neurons.

Natural and artificial audition can in principle acquire different solutions to a given problem. The constraints of the task, however, can nudge the cognitive science and engineering of audition to qualitatively converge, suggesting that a closer mutual examination would potentially enrich artificial hearing systems and process models of the mind and brain. Speech recognition - an area ripe for such exploration - is inherently robust in humans to a number transformations at various spectrotemporal granularities. To what extent are these robustness profiles accounted for by high-performing neural network systems? We bring together experiments in speech recognition under a single synthesis framework to evaluate state-of-the-art neural networks as stimulus-computable, optimized observers. In a series of experiments, we (1) clarify how influential speech manipulations in the literature relate to each other and to natural speech, (2) show the granularities at which machines exhibit out-of-distribution robustness, reproducing classical perceptual phenomena in humans, (3) identify the specific conditions where model predictions of human performance differ, and (4) demonstrate a crucial failure of all artificial systems to perceptually recover where humans do, suggesting alternative directions for theory and model building. These findings encourage a tighter synergy between the cognitive science and engineering of audition.

Convolutional neural networks (CNNs) are often described as promising models of human vision, yet they show many differences from human abilities. We focus on a superhuman capacity of top-performing CNNs, namely, their ability to learn very large datasets of random patterns. We verify that human learning on such tasks is extremely limited, even with few stimuli. We argue that the performance difference is due to CNNs' overcapacity and introduce biologically inspired mechanisms to constrain it, while retaining the good test set generalisation to structured images as characteristic of CNNs. We investigate the efficacy of adding noise to hidden units' activations, restricting early convolutional layers with a bottleneck, and using a bounded activation function. Internal noise was the most potent intervention and the only one which, by itself, could reduce random data performance in the tested models to chance levels. We also investigated whether networks with biologically inspired capacity constraints show improved generalisation to out-of-distribution stimuli, however little benefit was observed. Our results suggest that constraining networks with biologically motivated mechanisms paves the way for closer correspondence between network and human performance, but the few manipulations we have tested are only a small step towards that goal.

Recent research has shown that generative models with highly disentangled representations fail to generalise to unseen combination of generative factor values. These findings contradict earlier research which showed improved performance in out-of-training distribution settings when compared to entangled representations. Additionally, it is not clear if the reported failures are due to (a) encoders failing to map novel combinations to the proper regions of the latent space, or (b) novel combinations being mapped correctly but the decoder is unable to render the correct output for the unseen combinations. We investigate these alternatives by testing several models on a range of datasets and training settings. We find that (i) when models fail, their encoders also fail to map unseen combinations to correct regions of the latent space and (ii) when models succeed, it is either because the test conditions do not exclude enough examples, or because excluded cases involve combinations of object properties with it's shape. We argue that to generalise properly, models not only need to capture factors of variation, but also understand how to invert the process that causes the visual stimulus.

Humans rely heavily on the shape of objects to recognise them. Recently, it has been argued that Convolutional Neural Networks (CNNs) can also show a shape-bias, provided their learning environment contains this bias. This has led to the proposal that CNNs provide good mechanistic models of shape-bias and, more generally, human visual processing. However, it is also possible that humans and CNNs show a shape-bias for very different reasons, namely, shape-bias in humans may be a consequence of architectural and cognitive constraints whereas CNNs show a shape-bias as a consequence of learning the statistics of the environment. We investigated this question by exploring shape-bias in humans and CNNs when they learn in a novel environment. We observed that, in this new environment, humans (i) focused on shape and overlooked many non-shape features, even when non-shape features were more diagnostic, (ii) learned based on only one out of multiple predictive features, and (iii) failed to learn when global features, such as shape, were absent. This behaviour contrasted with the predictions of a statistical inference model with no priors, showing the strong role that shape-bias plays in human feature selection. It also contrasted with CNNs that (i) preferred to categorise objects based on non-shape features, and (ii) increased reliance on these non-shape features as they became more predictive. This was the case even when the CNN was pre-trained to have a shape-bias and the convolutional backbone was frozen. These results suggest that shape-bias has a different source in humans and CNNs: while learning in CNNs is driven by the statistical properties of the environment, humans are highly constrained by their previous biases, which suggests that cognitive constraints play a key role in how humans learn to recognise novel objects.

When seeing a new object, humans can immediately recognize it across different retinal locations: the internal object representation is invariant to translation. It is commonly believed that Convolutional Neural Networks (CNNs) are architecturally invariant to translation thanks to the convolution and/or pooling operations they are endowed with. In fact, several studies have found that these networks systematically fail to recognise new objects on untrained locations. In this work, we test a wide variety of CNNs architectures showing how, apart from DenseNet-121, none of the models tested was architecturally invariant to translation. Nevertheless, all of them could learn to be invariant to translation. We show how this can be achieved by pretraining on ImageNet, and it is sometimes possible with much simpler data sets when all the items are fully translated across the input canvas. At the same time, this invariance can be disrupted by further training due to catastrophic forgetting/interference. These experiments show how pretraining a network on an environment with the right 'latent' characteristics (a more naturalistic environment) can result in the network learning deep perceptual rules which would dramatically improve subsequent generalization.

Recently, domain-general recurrent neural networks, without explicit linguistic inductive biases, have been shown to successfully reproduce a range of human language behaviours, such as accurately predicting number agreement between nouns and verbs.We show that such networks will also learn number agreement within unnatural sentence structures, i.e. structures that are not found within any natural languages and which humans struggle to process.These results suggest that the models are learning from their input in a manner that is substantially different from human language acquisition, and we undertake an analysis of how the learned knowledge is stored in the weights of the network.We find that while the model has an effective understanding of singular versus plural for individual sentences, there is a lack of a unified concept of number agreement connecting these processes across the full range of inputs.Moreover, the weights handling natural and unnatural structures overlap substantially, in a way that underlines the non-human-like nature of the knowledge learned by the network.

We argue that symmetry is an important consideration in addressing the problem of systematic generalisation and investigate two forms of symmetry relevant to symbolic processes. We implement this approach in terms of convolution and show that it can be used to achieve effective generalisation in a rule learning and a context free language task.In the rule learning task, we find that symmetry allows us to learn rules that abstract away from the particular symbols that instantiate them, enabling generalisation from seen to unseen symbols. In the language task, symmetry allows us to impose a stack like architecture on the memory cells of a recurrent net, which permits generalisation from simple to more complex structures.

The key claim associated with a grandmother cell theory is that single neurons selectively represent one complex ‘thing’ (e.g. object and face). However, this theory is often mischaracterised in the cognitive and neuroscience literatures. I summarise two common confusions here. First, critics of grandmother cells often fail to distinguish between the selectivity and sparseness of neural firing and, as a result, predict (incorrectly) that one and only one neuron should fire in response to a given input. Second, critics often fail to distinguish between what a neuron responds to and what it represents – as detailed below – and as a result, predict (incorrectly) that a grandmother cell should fire in response to one and only one thing. I argue that these two confusions often lead to the premature rejection of grandmother cell theories.