Author name cluster

Stanley J. Osher

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

7 papers

2 author rows

ICLR Conference 2023 Conference Paper

A Primal-Dual Framework for Transformers and Neural Networks

Tan Minh Nguyen
Tam Minh Nguyen
Nhat Ho
Andrea L. Bertozzi
Richard G. Baraniuk
Stanley J. Osher

Self-attention is key to the remarkable success of transformers in sequence modeling tasks including many applications in natural language processing and computer vision. Like neural network layers, these attention mechanisms are often developed by heuristics and experience. To provide a principled framework for constructing attention layers in transformers, we show that the self-attention corresponds to the support vector expansion derived from a support vector regression problem, whose primal formulation has the form of a neural network layer. Using our framework, we derive popular attention layers used in practice and propose two new attentions: 1) the Batch Normalized Attention (Attention-BN) derived from the batch normalization layer and 2) the Attention with Scaled Head (Attention-SH) derived from using less training data to fit the SVR model. We empirically demonstrate the advantages of the Attention-BN and Attention-SH in reducing head redundancy, increasing the model's accuracy, and improving the model's efficiency in a variety of practical applications including image and time-series classification.

Details

ICML Conference 2023 Conference Paper

Neural Collapse in Deep Linear Networks: From Balanced to Imbalanced Data

Hien Dang 0003
Tho Tran Huu
Stanley J. Osher
Hung Tran-The
Nhat Ho
Tan Minh Nguyen

Modern deep neural networks have achieved impressive performance on tasks from image classification to natural language processing. Surprisingly, these complex systems with massive amounts of parameters exhibit the same structural properties in their last-layer features and classifiers across canonical datasets when training until convergence. In particular, it has been observed that the last-layer features collapse to their class-means, and those class-means are the vertices of a simplex Equiangular Tight Frame (ETF). This phenomenon is known as Neural Collapse (NC). Recent papers have theoretically shown that NC emerges in the global minimizers of training problems with the simplified “unconstrained feature model”. In this context, we take a step further and prove the NC occurrences in deep linear networks for the popular mean squared error (MSE) and cross entropy (CE) losses, showing that global solutions exhibit NC properties across the linear layers. Furthermore, we extend our study to imbalanced data for MSE loss and present the first geometric analysis of NC under bias-free setting. Our results demonstrate the convergence of the last-layer features and classifiers to a geometry consisting of orthogonal vectors, whose lengths depend on the amount of data in their corresponding classes. Finally, we empirically validate our theoretical analyses on synthetic and practical network architectures with both balanced and imbalanced scenarios.

Details

ICML Conference 2023 Conference Paper

Revisiting Over-smoothing and Over-squashing Using Ollivier-Ricci Curvature

Khang Nguyen 0002
Nong Minh Hieu
Vinh Duc Nguyen
Nhat Ho
Stanley J. Osher
Tan Minh Nguyen

Graph Neural Networks (GNNs) had been demonstrated to be inherently susceptible to the problems of over-smoothing and over-squashing. These issues prohibit the ability of GNNs to model complex graph interactions by limiting their effectiveness in taking into account distant information. Our study reveals the key connection between the local graph geometry and the occurrence of both of these issues, thereby providing a unified framework for studying them at a local scale using the Ollivier-Ricci curvature. Specifically, we demonstrate that over-smoothing is linked to positive graph curvature while over-squashing is linked to negative graph curvature. Based on our theory, we propose the Batch Ollivier-Ricci Flow, a novel rewiring algorithm capable of simultaneously addressing both over-smoothing and over-squashing.

Details

ICLR Conference 2022 Conference Paper

GRAND++: Graph Neural Diffusion with A Source Term

Matthew Thorpe
Tan Minh Nguyen
Hedi Xia
Thomas Strohmer
Andrea L. Bertozzi
Stanley J. Osher
Bao Wang 0001

We propose GRAph Neural Diffusion with a source term (GRAND++) for graph deep learning with a limited number of labeled nodes, i.e., low-labeling rate. GRAND++ is a class of continuous-depth graph deep learning architectures whose theoretical underpinning is the diffusion process on graphs with a source term. The source term guarantees two interesting theoretical properties of GRAND++: (i) the representation of graph nodes, under the dynamics of GRAND++, will not converge to a constant vector over all nodes even as the time goes to infinity, which mitigates the over-smoothing issue of graph neural networks and enables graph learning in very deep architectures. (ii) GRAND++ can provide accurate classification even when the model is trained with a very limited number of labeled training data. We experimentally verify the above two advantages on various graph deep learning benchmark tasks, showing a significant improvement over many existing graph neural networks.

Details

ICML Conference 2022 Conference Paper

Improving Transformers with Probabilistic Attention Keys

Tam Minh Nguyen
Tan Minh Nguyen
Dung D. Le
Duy Khuong Nguyen
Viet-Anh Tran
Richard G. Baraniuk
Nhat Ho
Stanley J. Osher

Multi-head attention is a driving force behind state-of-the-art transformers, which achieve remarkable performance across a variety of natural language processing (NLP) and computer vision tasks. It has been observed that for many applications, those attention heads learn redundant embedding, and most of them can be removed without degrading the performance of the model. Inspired by this observation, we propose Transformer with a Mixture of Gaussian Keys (Transformer-MGK), a novel transformer architecture that replaces redundant heads in transformers with a mixture of keys at each head. These mixtures of keys follow a Gaussian mixture model and allow each attention head to focus on different parts of the input sequence efficiently. Compared to its conventional transformer counterpart, Transformer-MGK accelerates training and inference, has fewer parameters, and requires fewer FLOPs to compute while achieving comparable or better accuracy across tasks. Transformer-MGK can also be easily extended to use with linear attention. We empirically demonstrate the advantage of Transformer-MGK in a range of practical applications, including language modeling and tasks that involve very long sequences. On the Wikitext-103 and Long Range Arena benchmark, Transformer-MGKs with 4 heads attain comparable or better performance to the baseline transformers with 8 heads.

Details

JMLR Journal 2014 Journal Article

Optimal Data Collection For Informative Rankings Expose Well-Connected Graphs

Braxton Osting
Christoph Brune
Stanley J. Osher

Given a graph where vertices represent alternatives and arcs represent pairwise comparison data, the statistical ranking problem is to find a potential function, defined on the vertices, such that the gradient of the potential function agrees with the pairwise comparisons. Our goal in this paper is to develop a method for collecting data for which the least squares estimator for the ranking problem has maximal Fisher information. Our approach, based on experimental design, is to view data collection as a bi-level optimization problem where the inner problem is the ranking problem and the outer problem is to identify data which maximizes the informativeness of the ranking. Under certain assumptions, the data collection problem decouples, reducing to a problem of finding multigraphs with large algebraic connectivity. This reduction of the data collection problem to graph-theoretic questions is one of the primary contributions of this work. As an application, we study the Yahoo! Movie user rating data set and demonstrate that the addition of a small number of well-chosen pairwise comparisons can significantly increase the Fisher informativeness of the ranking. As another application, we study the 2011-12 NCAA football schedule and propose schedules with the same number of games which are significantly more informative. Using spectral clustering methods to identify highly-connected communities within the division, we argue that the NCAA could improve its notoriously poor rankings by simply scheduling more out-of- conference games. [abs] [ pdf ][ bib ] &copy JMLR 2014. ( edit, beta )

PDF Details

ICML Conference 2013 Conference Paper

Enhanced statistical rankings via targeted data collection

Braxton Osting
Christoph Brune
Stanley J. Osher

Given a graph where vertices represent alternatives and pairwise comparison data, y_ij, is given on the edges, the statistical ranking problem is to find a potential function, defined on the vertices, such that the gradient of the potential function agrees with pairwise comparisons. We study the dependence of the statistical ranking problem on the available pairwise data, i. e. , pairs (i, j) for which the pairwise comparison data y_ij is known, and propose a framework to identify data which, when augmented with the current dataset, maximally increases the Fisher information of the ranking. Under certain assumptions, the data collection problem decouples, reducing to a problem of finding an edge set on the graph (with a fixed number of edges) such that the second eigenvalue of the graph Laplacian is maximal. This reduction of the data collection problem to a spectral graph-theoretic question is one of the primary contributions of this work. As an application, we study the Yahoo! Movie user rating dataset and demonstrate that the addition of a small number of well-chosen pairwise comparisons can significantly increase the Fisher informativeness of the ranking.

Details