arXiv subjects

Tomaso Poggio

Publications and source records attributed to Tomaso Poggio.

At least 55 records · Page 3Linked to original sources

arXiv2023

Norm-based Generalization Bounds for Compositionally Sparse Neural Networks

In this paper, we investigate the Rademacher complexity of deep sparse neural networks, where each neuron receives a small number of inputs. We prove generalization bounds for multilayered sparse ReLU neural networks, including convolutional neural networks. These bounds differ from previous ones, as they consider the norms of the convolutional filters instead of the norms of the associated Toeplitz matrices, independently of weight sharing between neurons. As we show theoretically, these bounds may be orders of magnitude better than standard norm-based generalization bounds and empirically, they are almost non-vacuous in estimating generalization in various simple classification problems. Taken together, these results suggest that compositional sparsity of the underlying target function is critical to the success of deep neural networks.

cs.LG↗

arXiv2021

Neural-guided, Bidirectional Program Search for Abstraction and Reasoning

One of the challenges facing artificial intelligence research today is designing systems capable of utilizing systematic reasoning to generalize to new tasks. The Abstraction and Reasoning Corpus (ARC) measures such a capability through a set of visual reasoning tasks. In this paper we report incremental progress on ARC and lay the foundations for two approaches to abstraction and reasoning not based in brute-force search. We first apply an existing program synthesis system called DreamCoder to create symbolic abstractions out of tasks solved so far, and show how it enables solving of progressively more challenging ARC tasks. Second, we design a reasoning algorithm motivated by the way humans approach ARC. Our algorithm constructs a search graph and reasons over this graph structure to discover task solutions. More specifically, we extend existing execution-guided program synthesis approaches with deductive reasoning based on function inverse semantics to enable a neural-guided bidirectional search algorithm. We demonstrate the effectiveness of the algorithm on three domains: ARC, 24-Game tasks, and a 'double-and-add' arithmetic puzzle.

cs.AI↗

arXiv2021

Distribution of Classification Margins: Are All Data Equal?

Recent theoretical results show that gradient descent on deep neural networks under exponential loss functions locally maximizes classification margin, which is equivalent to minimizing the norm of the weight matrices under margin constraints. This property of the solution however does not fully characterize the generalization performance. We motivate theoretically and show empirically that the area under the curve of the margin distribution on the training set is in fact a good measure of generalization. We then show that, after data separation is achieved, it is possible to dynamically reduce the training set by more than 99% without significant loss of performance. Interestingly, the resulting subset of "high capacity" features is not consistent across different training runs, which is consistent with the theoretical claim that all training points should converge to the same asymptotic margin under SGD and in the presence of both batch normalization and weight decay.

cs.LG↗

arXiv2021

Hierarchically Compositional Tasks and Deep Convolutional Networks

The main success stories of deep learning, starting with ImageNet, depend on deep convolutional networks, which on certain tasks perform significantly better than traditional shallow classifiers, such as support vector machines, and also better than deep fully connected networks; but what is so special about deep convolutional networks? Recent results in approximation theory proved an exponential advantage of deep convolutional networks with or without shared weights in approximating functions with hierarchical locality in their compositional structure. More recently, the hierarchical structure was proved to be hard to learn from data, suggesting that it is a powerful prior embedded in the architecture of the network. These mathematical results, however, do not say which real-life tasks correspond to input-output functions with hierarchical locality. To evaluate this, we consider a set of visual tasks where we disrupt the local organization of images via "deterministic scrambling" to later perform a visual task on these images structurally-altered in the same way for training and testing. For object recognition we find, as expected, that scrambling does not affect the performance of shallow or deep fully connected networks contrary to the out-performance of convolutional networks. Not all tasks involving images are however affected. Texture perception and global color estimation are much less sensitive to deterministic scrambling showing that the underlying functions corresponding to these tasks are not hierarchically local; and also counter-intuitively showing that these tasks are better approximated by networks that are not deep (texture) nor convolutional (color). Altogether, these results shed light into the importance of matching a network architecture with its embedded prior of the task to be learned.

cs.LG↗

arXiv2021

The Effects of Image Distribution and Task on Adversarial Robustness

In this paper, we propose an adaptation to the area under the curve (AUC) metric to measure the adversarial robustness of a model over a particular $ε$-interval $[ε_0, ε_1]$ (interval of adversarial perturbation strengths) that facilitates unbiased comparisons across models when they have different initial $ε_0$ performance. This can be used to determine how adversarially robust a model is to different image distributions or task (or some other variable); and/or to measure how robust a model is comparatively to other models. We used this adversarial robustness metric on models of an MNIST, CIFAR-10, and a Fusion dataset (CIFAR-10 + MNIST) where trained models performed either a digit or object recognition task using a LeNet, ResNet50, or a fully connected network (FullyConnectedNet) architecture and found the following: 1) CIFAR-10 models are inherently less adversarially robust than MNIST models; 2) Both the image distribution and task that a model is trained on can affect the adversarial robustness of the resultant model. 3) Pretraining with a different image distribution and task sometimes carries over the adversarial robustness induced by that image distribution and task in the resultant model; Collectively, our results imply non-trivial differences of the learned representation space of one perceptual system over another given its exposure to different image statistics or tasks (mainly objects vs digits). Moreover, these results hold even when model systems are equalized to have the same level of performance, or when exposed to approximately matched image statistics of fusion images but with different tasks.

cs.LG↗

arXiv2020

Bridging the Gaps Between Residual Learning, Recurrent Neural Networks and Visual Cortex

We discuss relations between Residual Networks (ResNet), Recurrent Neural Networks (RNNs) and the primate visual cortex. We begin with the observation that a special type of shallow RNN is exactly equivalent to a very deep ResNet with weight sharing among the layers. A direct implementation of such a RNN, although having orders of magnitude fewer parameters, leads to a performance similar to the corresponding ResNet. We propose 1) a generalization of both RNN and ResNet architectures and 2) the conjecture that a class of moderately deep RNNs is a biologically-plausible model of the ventral stream in visual cortex. We demonstrate the effectiveness of the architectures by testing them on the CIFAR-10 and ImageNet dataset.

cs.LG↗

arXiv2020

Explicit regularization and implicit bias in deep network classifiers trained with the square loss

Deep ReLU networks trained with the square loss have been observed to perform well in classification tasks. We provide here a theoretical justification based on analysis of the associated gradient flow. We show that convergence to a solution with the absolute minimum norm is expected when normalization techniques such as Batch Normalization (BN) or Weight Normalization (WN) are used together with Weight Decay (WD). The main property of the minimizers that bounds their expected error is the norm: we prove that among all the close-to-interpolating solutions, the ones associated with smaller Frobenius norms of the unnormalized weight matrices have better margin and better bounds on the expected classification error. With BN but in the absence of WD, the dynamical system is singular. Implicit dynamical regularization -- that is zero-initial conditions biasing the dynamics towards high margin solutions -- is also possible in the no-BN and no-WD case. The theory yields several predictions, including the role of BN and weight decay, aspects of Papyan, Han and Donoho's Neural Collapse and the constraints induced by BN on the network weights.

cs.LG↗

arXiv2020

CUDA-Optimized real-time rendering of a Foveated Visual System

The spatially-varying field of the human visual system has recently received a resurgence of interest with the development of virtual reality (VR) and neural networks. The computational demands of high resolution rendering desired for VR can be offset by savings in the periphery, while neural networks trained with foveated input have shown perceptual gains in i.i.d and o.o.d generalization. In this paper, we present a technique that exploits the CUDA GPU architecture to efficiently generate Gaussian-based foveated images at high definition (1920x1080 px) in real-time (165 Hz), with a larger number of pooling regions than previous Gaussian-based foveation algorithms by several orders of magnitude, producing a smoothly foveated image that requires no further blending or stitching, and that can be well fit for any contrast sensitivity function. The approach described can be adapted from Gaussian blurring to any eccentricity-dependent image processing and our algorithm can meet demand for experimentation to evaluate the role of spatially-varying processing across biological and artificial agents, so that foveation can be added easily on top of existing systems rather than forcing their redesign (emulated foveated renderer). Altogether, this paper demonstrates how a GPU, with a CUDA block-wise architecture, can be employed for radially-variant rendering, with opportunities for more complex post-processing to ensure a metameric foveation scheme. Code is provided.

eess.IV↗

arXiv2020

For interpolating kernel machines, minimizing the norm of the ERM solution minimizes stability

We study the average $\mbox{CV}_{loo}$ stability of kernel ridge-less regression and derive corresponding risk bounds. We show that the interpolating solution with minimum norm minimizes a bound on $\mbox{CV}_{loo}$ stability, which in turn is controlled by the condition number of the empirical kernel matrix. The latter can be characterized in the asymptotic regime where both the dimension and cardinality of the data go to infinity. Under the assumption of random kernel matrices, the corresponding test error should be expected to follow a double descent curve.

stat.ML↗

arXiv2020

Biologically Inspired Mechanisms for Adversarial Robustness

A convolutional neural network strongly robust to adversarial perturbations at reasonable computational and performance cost has not yet been demonstrated. The primate visual ventral stream seems to be robust to small perturbations in visual stimuli but the underlying mechanisms that give rise to this robust perception are not understood. In this work, we investigate the role of two biologically plausible mechanisms in adversarial robustness. We demonstrate that the non-uniform sampling performed by the primate retina and the presence of multiple receptive fields with a range of receptive field sizes at each eccentricity improve the robustness of neural networks to small adversarial perturbations. We verify that these two mechanisms do not suffer from gradient obfuscation and study their contribution to adversarial robustness through ablation studies.

cs.LG↗

arXiv2020

Double descent in the condition number

In solving a system of $n$ linear equations in $d$ variables $Ax=b$, the condition number of the $n,d$ matrix $A$ measures how much errors in the data $b$ affect the solution $x$. Estimates of this type are important in many inverse problems. An example is machine learning where the key task is to estimate an underlying function from a set of measurements at random points in a high dimensional space and where low sensitivity to error in the data is a requirement for good predictive performance. Here we discuss the simple observation, which is known but surprisingly little quoted (see Theorem 4.2 in \cite{Brgisser:2013:CGN:2526261}): when the columns of $A$ are random vectors, the condition number of $A$ is highest if $d=n$, that is when the inverse of $A$ exists. An overdetermined system ($n>d$) as well as an underdetermined system ($n<d$), for which the pseudoinverse must be used instead of the inverse, typically have significantly better, that is lower, condition numbers. Thus the condition number of $A$ plotted as function of $d$ shows a double descent behavior with a peak at $d=n$.

cs.LG↗

arXiv2020

Theory III: Dynamics and Generalization in Deep Networks

The key to generalization is controlling the complexity of the network. However, there is no obvious control of complexity -- such as an explicit regularization term -- in the training of deep networks for classification. We will show that a classical form of norm control -- but kind of hidden -- is present in deep networks trained with gradient descent techniques on exponential-type losses. In particular, gradient descent induces a dynamics of the normalized weights which converge for $t \to \infty$ to an equilibrium which corresponds to a minimum norm (or maximum margin) solution. For sufficiently large but finite $ρ$ -- and thus finite $t$ -- the dynamics converges to one of several margin maximizers, with the margin monotonically increasing towards a limit stationary point of the flow. In the usual case of stochastic gradient descent, most of the stationary points are likely to be convex minima corresponding to a constrained minimizer -- the network with normalized weights-- which corresponds to vanishing regularization. The solution has zero generalization gap, for fixed architecture, asymptotically for $N \to \infty$, where $N$ is the number of training examples. Our approach extends some of the original results of Srebro from linear networks to deep networks and provides a new perspective on the implicit bias of gradient descent. We believe that the elusive complexity control we describe is responsible for the puzzling empirical finding of good predictive performance by deep networks, despite overparametrization.

cs.LG↗

arXiv2019

An analysis of training and generalization errors in shallow and deep networks

This paper is motivated by an open problem around deep networks, namely, the apparent absence of over-fitting despite large over-parametrization which allows perfect fitting of the training data. In this paper, we analyze this phenomenon in the case of regression problems when each unit evaluates a periodic activation function. We argue that the minimal expected value of the square loss is inappropriate to measure the generalization error in approximation of compositional functions in order to take full advantage of the compositional structure. Instead, we measure the generalization error in the sense of maximum loss, and sometimes, as a pointwise error. We give estimates on exactly how many parameters ensure both zero training error as well as a good generalization error. We prove that a solution of a regularization problem is guaranteed to yield a good training error as well as a good generalization error and estimate how much error to expect at which test data.

cs.LG↗

arXiv2019

Theoretical Issues in Deep Networks: Approximation, Optimization and Generalization

While deep learning is successful in a number of applications, it is not yet well understood theoretically. A satisfactory theoretical characterization of deep learning however, is beginning to emerge. It covers the following questions: 1) representation power of deep networks 2) optimization of the empirical risk 3) generalization properties of gradient descent techniques --- why the expected error does not suffer, despite the absence of explicit regularization, when the networks are overparametrized? In this review we discuss recent advances in the three areas. In approximation theory both shallow and deep networks have been shown to approximate any continuous functions on a bounded domain at the expense of an exponential number of parameters (exponential in the dimensionality of the function). However, for a subset of compositional functions, deep networks of the convolutional type can have a linear dependence on dimensionality, unlike shallow networks. In optimization we discuss the loss landscape for the exponential loss function and show that stochastic gradient descent will find with high probability the global minima. To address the question of generalization for classification tasks, we use classical uniform convergence results to justify minimizing a surrogate exponential-type loss function under a unit norm constraint on the weight matrix at each layer -- since the interesting variables for classification are the weight directions rather than the weights. Our approach, which is supported by several independent new results, offers a solution to the puzzle about generalization performance of deep overparametrized ReLU networks, uncovering the origin of the underlying hidden complexity control.

cs.LG↗

arXiv2019

Deep Recurrent Architectures for Seismic Tomography

This paper introduces novel deep recurrent neural network architectures for Velocity Model Building (VMB), which is beyond what Araya-Polo et al 2018 pioneered with the Machine Learning-based seismic tomography built with convolutional non-recurrent neural network. Our investigation includes the utilization of basic recurrent neural network (RNN) cells, as well as Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) cells. Performance evaluation reveals that salt bodies are consistently predicted more accurately by GRU and LSTM-based architectures, as compared to non-recurrent architectures. The results take us a step closer to the final goal of a reliable fully Machine Learning-based tomography from pre-stack data, which when achieved will reduce the VMB turnaround from weeks to days.

physics.geo-ph↗

arXiv2019

Fisher-Rao Metric, Geometry, and Complexity of Neural Networks

We study the relationship between geometry and capacity measures for deep neural networks from an invariance viewpoint. We introduce a new notion of capacity --- the Fisher-Rao norm --- that possesses desirable invariance properties and is motivated by Information Geometry. We discover an analytical characterization of the new capacity measure, through which we establish norm-comparison inequalities and further show that the new measure serves as an umbrella for several existing norm-based complexity measures. We discuss upper bounds on the generalization error induced by the proposed measure. Extensive numerical experiments on CIFAR-10 support our theoretical findings. Our theoretical analysis rests on a key structural lemma about partial derivatives of multi-layer rectifier networks.

cs.LG↗

arXiv2018

Biologically-plausible learning algorithms can scale to large datasets

The backpropagation (BP) algorithm is often thought to be biologically implausible in the brain. One of the main reasons is that BP requires symmetric weight matrices in the feedforward and feedback pathways. To address this "weight transport problem" (Grossberg, 1987), two more biologically plausible algorithms, proposed by Liao et al. (2016) and Lillicrap et al. (2016), relax BP's weight symmetry requirements and demonstrate comparable learning capabilities to that of BP on small datasets. However, a recent study by Bartunov et al. (2018) evaluate variants of target-propagation (TP) and feedback alignment (FA) on MINIST, CIFAR, and ImageNet datasets, and find that although many of the proposed algorithms perform well on MNIST and CIFAR, they perform significantly worse than BP on ImageNet. Here, we additionally evaluate the sign-symmetry algorithm (Liao et al., 2016), which differs from both BP and FA in that the feedback and feedforward weights share signs but not magnitudes. We examine the performance of sign-symmetry and feedback alignment on ImageNet and MS COCO datasets using different network architectures (ResNet-18 and AlexNet for ImageNet, RetinaNet for MS COCO). Surprisingly, networks trained with sign-symmetry can attain classification performance approaching that of BP-trained networks. These results complement the study by Bartunov et al. (2018), and establish a new benchmark for future biologically plausible learning algorithms on more difficult datasets and more complex architectures.

cs.LG↗

arXiv2018

A Surprising Linear Relationship Predicts Test Performance in Deep Networks

Given two networks with the same training loss on a dataset, when would they have drastically different test losses and errors? Better understanding of this question of generalization may improve practical applications of deep networks. In this paper we show that with cross-entropy loss it is surprisingly simple to induce significantly different generalization performances for two networks that have the same architecture, the same meta parameters and the same training error: one can either pretrain the networks with different levels of "corrupted" data or simply initialize the networks with weights of different Gaussian standard deviations. A corollary of recent theoretical results on overfitting shows that these effects are due to an intrinsic problem of measuring test performance with a cross-entropy/exponential-type loss, which can be decomposed into two components both minimized by SGD -- one of which is not related to expected classification performance. However, if we factor out this component of the loss, a linear relationship emerges between training and test losses. Under this transformation, classical generalization bounds are surprisingly tight: the empirical/training loss is very close to the expected/test loss. Furthermore, the empirical relation between classification error and normalized cross-entropy loss seem to be approximately monotonic

cs.LG↗