Abstract In this paper, we explore the structure of the penultimate Gram matrix in deep neural networks, which contains the pairwise inner products of outputs corresponding to a batch of inputs. In several architectures it has been observed that this Gram matrix becomes degenerate with depth at initialization, which dramatically slows training. Normalization layers, such as batch or layer normalization, play a pivotal role in preventing the rank collapse issue. Despite promising advances, the existing theoretical results (i) do not extend to layer normalization, which is widely used in transformers, (ii) can not characterize the bias of normalization quantitatively at finite depth. To bridge this gap, we provide a proof that layer normalization, in conjunction with activation layers, biases the Gram matrix of a multilayer perceptron towards isometry at an exponential rate with depth at initialization. We quantify this rate using the Hermite expansion of the activation function, highlighting the importance of higher order (≥2) Hermite coefficients in the bias towards isometry.

Authors Amir Joudaki, Hadi Daneshmand, Francis Bach

Submitted NeurIPS 2023 (poster)

Abstract Selecting hyperparameters in deep learning greatly impacts its effectiveness but requires manual effort and expertise. Recent works show that Bayesian model selection with Laplace approximations can allow to optimize such hyperparameters just like standard neural network parameters using gradients and on the training data. However, estimating a single hyperparameter gradient requires a pass through the entire dataset, limiting the scalability of such algorithms. In this work, we overcome this issue by introducing lower bounds to the linearized Laplace approximation of the marginal likelihood. In contrast to previous estimators, these bounds are amenable to stochastic-gradient-based optimization and allow to trade off estimation accuracy against computational complexity. We derive them using the function-space form of the linearized Laplace, which can be estimated using the neural tangent kernel. Experimentally, we show that the estimators can significantly accelerate gradient-based hyperparameter optimization.

Authors Alexander Immer, Tycho FA van der Ouderaa, Mark van der Wilk, Gunnar Rätsch, Bernhard Schölkopf

Submitted ICML 2023

Link

Abstract Mean-field theory is widely used in theoretical studies of neural networks. In this paper, we analyze the role of depth in the concentration of mean-field predictions for Gram matrices of hidden representations in deep multilayer perceptron (MLP) with batch normalization (BN) at initialization. It is postulated that the mean-field predictions suffer from layer-wise errors that amplify with depth. We demonstrate that BN avoids this error amplification with depth. When the chain of hidden representations is rapidly mixing, we establish a concentration bound for a mean-field model of Gram matrices. To our knowledge, this is the first concentration bound that does not become vacuous with depth for standard MLPs with a finite width.

Authors Amir Joudaki, Hadi Daneshmand, Francis Bach

Submitted ICML 2023 (poster)

Link

Abstract Models that can predict the occurrence of events ahead of time with low false-alarm rates are critical to the acceptance of decision support systems in the medical community. This challenging task is typically treated as a simple binary classification, ignoring temporal dependencies between samples, whereas we propose to exploit this structure. We first introduce a common theoretical framework unifying dynamic survival analysis and early event prediction. Following an analysis of objectives from both fields, we propose Temporal Label Smoothing (TLS), a simpler, yet best-performing method that preserves prediction monotonicity over time. By focusing the objective on areas with a stronger predictive signal, TLS improves performance over all baselines on two large-scale benchmark tasks. Gains are particularly notable along clinically relevant measures, such as event recall at low false-alarm rates. TLS reduces the number of missed events by up to a factor of two over previously used approaches in early event prediction.

Authors Hugo Yèche, Alizée Pace, Gunnar Rätsch, Rita Kuznetsova

Submitted ICML 2023

Link DOI

Abstract Understanding and predicting molecular responses towards external perturbations is a core question in molecular biology. Technological advancements in the recent past have enabled the generation of high-resolution single-cell data, making it possible to profile individual cells under different experimentally controlled perturbations. However, cells are typically destroyed during measurement, resulting in unpaired distributions over either perturbed or non-perturbed cells. Leveraging the theory of optimal transport and the recent advents of convex neural architectures, we learn a coupling describing the response of cell populations upon perturbation, enabling us to predict state trajectories on a single-cell level. We apply our approach, CellOT, to predict treatment responses of 21,650 cells subject to four different drug perturbations. CellOT outperforms current state-of-the-art methods both qualitatively and quantitatively, accurately capturing cellular behavior shifts across all different drugs.

Authors Charlotte Bunne, Stefan Stark, Gabriele Gut, Jacobo Sarabia del Castillo, Mitchell Levesque, Kjong Van Lehmann, Lucas Pelkmans, Andreas Krause, Gunnar Rätsch

Submitted BioRxiv

Link DOI