1 July 2024 in Physical Review X
Deep learning algorithms demonstrate a surprising ability to learn high-dimensional tasks from limited examples. This is commonly attributed to the depth of neural networks, enabling them to build a hierarchy of abstract, low-dimensional data representations. However, how many training examples are required to learn such representations remains unknown. To quantitatively study this question, we introduce the random hierarchy model: a family of synthetic tasks inspired by the hierarchical structure of language and images. The model is a classification task where each class corresponds to a group of high-level features, chosen among several equivalent groups associated with the same class. In turn, each feature corresponds to a group of subfeatures chosen among several equivalent groups and so on, following a hierarchy of composition rules. We find that deep networks learn the task by developing internal representations invariant to exchanging equivalent groups. Moreover, the number of data required corresponds to the point where correlations between low-level features and classes become detectable. Overall, our results indicate how deep networks overcome the curse of dimensionality by building invariant representations and provide an estimate of the number of data required to learn a hierarchical task. Published by the American Physical Society 2024
9 November 2023 in Machine Learning: Science and Technology
A central question of machine learning is how deep nets manage to learn tasks in high dimensions. An appealing hypothesis is that they achieve this feat by building a representation of the data where information irrelevant to the task is lost. For image datasets, this view is supported by the observation that after (and not before) training, the neural representation becomes less and less sensitive to diffeomorphisms acting on images as the signal propagates through the network. This loss of sensitivity correlates with performance and surprisingly correlates with a gain of sensitivity to white noise acquired during training. Which are the mechanisms learned by convolutional neural networks (CNNs) responsible for the these phenomena? In particular, why is the sensitivity to noise heightened with training? Our approach consists of two steps. (1) Analyzing the layer-wise representations of trained CNNs, we disentangle the role of spatial pooling in contrast to channel pooling in decreasing their sensitivity to image diffeomorphisms while increasing their sensitivity to noise. (2) We introduce model scale-detection tasks, which qualitatively reproduce the phenomena reported in our empirical analysis. In these models we can assess quantitatively how spatial pooling affects these sensitivities. We find that the increased sensitivity to noise observed in deep ReLU networks is a mechanistic consequence of the perturbing noise piling up during spatial pooling, after being rectified by ReLU units. Using odd activation functions like tanh drastically reduces the CNNs’ sensitivity to noise.
17 June 2021 in The European Physical Journal Special Topics
The goal of response theory, in each of its many statistical mechanical formulations, is to predict the perturbed response of a system from the knowledge of the unperturbed state and of the applied perturbation. A new recent angle on the problem focuses on providing a method to perform predictions of the change in one observable of the system using the change in a second observable as a surrogate for the actual forcing. Such a viewpoint tries to address the very relevant problem of causal links within complex system when only incomplete information is available. We present here a method for quantifying and ranking the predictive ability of observables and use it to investigate the response of a paradigmatic spatially extended system, the Lorenz ’96 model. We perturb locally the system and we then study to what extent a given local observable can predict the behaviour of a separate local observable. We show that this approach can reveal insights on the way a signal propagates inside the system. We also show that the procedure becomes more efficient if one considers multiple acting forcings and, correspondingly, multiple observables as predictors of the observable of interest.