One of the most impactful findings in computational neuroscience over the past decade is that the object recognition accuracy of deep neural networks (DNNs) correlates with their ability to predict neural responses to natural images in the inferotemporal (IT) cortex. This discovery supported the long-held theory that object recognition is a core objective of the visual cortex, and suggested that more accurate DNNs would serve as better models of IT neuron responses to images. Since then, deep learning has undergone a revolution of scale: billion parameter-scale DNNs trained on billions of images are rivaling or outperforming humans at visual tasks including object recognition. Have today's DNNs become more accurate at predicting IT neuron responses to images as they have grown more accurate at object recognition?Surprisingly, across three independent experiments, we find that this is not the case. DNNs have become progressively worse models of IT as their accuracy has increased on ImageNet.

One of the most impactful findings in computational neuroscience over the past decade is that the object recognition accuracy of deep neural networks (DNNs) correlates with their ability to predict neural responses to natural images in the inferotemporal (IT) cortex. This discovery supported the long-held theory that object recognition is a core objective of the visual cortex, and suggested that more accurate DNNs would serve as better models of IT neuron responses to images. Since then, deep learning has undergone a revolution of scale: billion parameter-scale DNNs trained on billions of images are rivaling or outperforming humans at visual tasks including object recognition. Have today's DNNs become more accurate at predicting IT neuron responses to images as they have grown more accurate at object recognition?Surprisingly, across three independent experiments, we find that this is not the case. DNNs have become progressively worse models of IT as their accuracy has increased on ImageNet.

This lack of transparency, often referred to as the "black box" problem, makes it hard to interpret the workings of learning-based robot control systems. Understanding why a particular decision was made or predicting how the system will behave in future scenarios remains a challenge, yet critical for physical deployments. Through the lens of representation learning, we assume that neural networks capture a set of processes that exist in the data distribution; for robots, they manifest learned skills, behaviors, or strategies, which are critical to understand the decision-making of a policy. However, while these factors of variation [1] (e.g., color or shape representations) are actively studied in unsupervised learning for disentangled representation, in robot learning, they are less well-defined and pose unique challenges due to the intertwined correspondence of neural activities with emergent behaviors unknown a priori. In the present study, we aim to (i) provide a useful definition of factors of variation for policy learning, and (ii) explore how to uncover dynamics and factors of variation quantitatively as a measure of interpretability in compact neural networks for closed-loop end-to-end control applica-7th Conference on Robot Learning (CoRL 2023), Atlanta, USA.

The mammalian brain could contain dense and sparse network connectivity structures, including both excitatory and inhibitory neurons, but is without any clearly defined output layer. The neurons have time constants, which mean that the integrated network structure has state memory. The network structure contains complex mutual interactions between the neurons under different conditions, which depend on the internal state of the network. The internal state can be defined as the distribution of activity across all individual neurons across the network. Therefore, the state of a neuron/network becomes a defining factor for how information is represented within the network. Towards this study, we constructed a fully connected (with dense/sparse coding strategies) recurrent network comprising of both excitatory and inhibitory neurons, driven by pseudo-random inputs of varying frequencies. In this study we assessed the impact of varying specific intrinsic parameters of the neurons that enriched network state dynamics, such as initial neuron activity, amount of inhibition in combination with thresholded neurons and conduction delays. The impact was assessed by quantifying the changes in mutual interactions between the neurons within the network for each given input. We found such effects were more profound in sparsely connected networks than in densely connected networks. However, also densely connected networks could make use of such dynamic changes in the mutual interactions between neurons, as a given input could induce multiple different network states.

An activation boundary for a neuron refers to a separating hyperplane that determines whether the neuron is activated or deactivated. It has been long considered in neural networks that the activations of neurons, rather than their exact output values, play the most important role in forming classification friendly partitions of the hidden feature space. However, as far as we know, this aspect of neural networks has not been considered in the literature of knowledge transfer. In this paper, we propose a knowledge transfer method via distillation of activation boundaries formed by hidden neurons. For the distillation, we propose an activation transfer loss that has the minimum value when the boundaries generated by the student coincide with those by the teacher. Since the activation transfer loss is not differentiable, we design a piecewise differentiable loss approximating the activation transfer loss. By the proposed method, the student learns a separating boundary between activation region and deactivation region formed by each neuron in the teacher. Through the experiments in various aspects of knowledge transfer, it is verified that the proposed method outperforms the current state-of-the-art.

You won't have seen it on the podium, but the human brain's mirror neuron system could have medaled at this year's Olympic Games, or basically any sporting event with an audience. The mirror neuron system is a network of neurons that activates both when you watch someone do something and when you do it yourself, and it turns out to be an important part of the subjective experience of being a fan. But watching a sport doesn't just flip your mirror neuron system on like a switch. There are degrees of activation. While you and the person sitting beside you probably both have your mirror neuron systems firing, your neighbor's neurons might have different levels of activation than yours.