Graph Transformers Dream of Electric Flow

We show theoretically and empirically that the linear Transformer, when applied to graph data, can implement algorithms that solve canonical problems such as electric flow and eigenvector decomposition. The input to the Transformer is simply the graph incidence matrix; no other explicit positional encoding information is provided. We present explicit weight configurations for implementing each such graph algorithm, and we bound the errors of the constructed Transformers by the errors of the underlying algorithms. Our theoretical findings are corroborated by experiments on synthetic data. Additionally, on a real-world molecular regression task, we observe that the linear Transformer is capable of learning a more effective positional encoding than the default one based on Laplacian eigenvectors. Our work is an initial step towards elucidating the inner-workings of the Transformer for graph data.

Transformer In-Context Learning for Categorical Data

Recent research has sought to understand Transformers through the lens of in-context learning with functional data. We extend that line of work with the goal of moving closer to language models, considering categorical outcomes, nonlinear underlying models, and nonlinear attention. The contextual data are of the form $\textsf{C}=(x_1,c_1,\dots,x_N,c_{N})$ where each $c_i\in\{0,\dots,C-1\}$ is drawn from a categorical distribution that depends on covariates $x_i\in\mathbb{R}^d$. Contextual outcomes in the $m$th set of contextual data, $\textsf{C}_m$, are modeled in terms of latent function $f_m(x)\in\textsf{F}$, where $\textsf{F}$ is a functional class with $(C-1)$-dimensional vector output. The probability of observing class $c\in\{0,\dots,C-1\}$ is modeled in terms of the output components of $f_m(x)$ via the softmax. The Transformer parameters may be trained with $M$ contextual examples, $\{\textsf{C}_m\}_{m=1,M}$, and the trained model is then applied to new contextual data $\textsf{C}_{M+1}$ for new $f_{M+1}(x)\in\textsf{F}$. The goal is for the Transformer to constitute the probability of each category $c\in\{0,\dots,C-1\}$ for a new query $x_{N_{M+1}+1}$. We assume each component of $f_m(x)$ resides in a reproducing kernel Hilbert space (RKHS), specifying $\textsf{F}$. Analysis and an extensive set of experiments suggest that on its forward pass the Transformer (with attention defined by the RKHS kernel) implements a form of gradient descent of the underlying function, connected to the latent vector function associated with the softmax. We present what is believed to be the first real-world demonstration of this few-shot-learning methodology, using the ImageNet dataset.

Meta-Learned Attribute Self-Interaction Network for Continual and Generalized Zero-Shot Learning

Zero-shot learning (ZSL) is a promising approach to generalizing a model to categories unseen during training by leveraging class attributes, but challenges remain. Recently, methods using generative models to combat bias towards classes seen during training have pushed state of the art, but these generative models can be slow or computationally expensive to train. Also, these generative models assume that the attribute vector of each unseen class is available a priori at training, which is not always practical. Additionally, while many previous ZSL methods assume a one-time adaptation to unseen classes, in reality, the world is always changing, necessitating a constant adjustment of deployed models. Models unprepared to handle a sequential stream of data are likely to experience catastrophic forgetting. We propose a Meta-learned Attribute self-Interaction Network (MAIN) for continual ZSL. By pairing attribute self-interaction trained using meta-learning with inverse regularization of the attribute encoder, we are able to outperform state-of-the-art results without leveraging the unseen class attributes while also being able to train our models substantially faster (>100x) than expensive generative-based approaches. We demonstrate this with experiments on five standard ZSL datasets (CUB, aPY, AWA1, AWA2, and SUN) in the generalized zero-shot learning and continual (fixed/dynamic) zero-shot learning settings. Extensive ablations and analyses demonstrate the efficacy of various components proposed.

Open World Classification with Adaptive Negative Samples

Open world classification is a task in natural language processing with key practical relevance and impact. Since the open or {\em unknown} category data only manifests in the inference phase, finding a model with a suitable decision boundary accommodating for the identification of known classes and discrimination of the open category is challenging. The performance of existing models is limited by the lack of effective open category data during the training stage or the lack of a good mechanism to learn appropriate decision boundaries. We propose an approach based on \underline{a}daptive \underline{n}egative \underline{s}amples (ANS) designed to generate effective synthetic open category samples in the training stage and without requiring any prior knowledge or external datasets. Empirically, we find a significant advantage in using auxiliary one-versus-rest binary classifiers, which effectively utilize the generated negative samples and avoid the complex threshold-seeking stage in previous works. Extensive experiments on three benchmark datasets show that ANS achieves significant improvements over state-of-the-art methods.

Pushing the Efficiency Limit Using Structured Sparse Convolutions

Weight pruning is among the most popular approaches for compressing deep convolutional neural networks. Recent work suggests that in a randomly initialized deep neural network, there exist sparse subnetworks that achieve performance comparable to the original network. Unfortunately, finding these subnetworks involves iterative stages of training and pruning, which can be computationally expensive. We propose Structured Sparse Convolution (SSC), which leverages the inherent structure in images to reduce the parameters in the convolutional filter. This leads to improved efficiency of convolutional architectures compared to existing methods that perform pruning at initialization. We show that SSC is a generalization of commonly used layers (depthwise, groupwise and pointwise convolution) in ``efficient architectures.'' Extensive experiments on well-known CNN models and datasets show the effectiveness of the proposed method. Architectures based on SSC achieve state-of-the-art performance compared to baselines on CIFAR-10, CIFAR-100, Tiny-ImageNet, and ImageNet classification benchmarks.

Pseudo-OOD training for robust language models

While pre-trained large-scale deep models have garnered attention as an important topic for many downstream natural language processing (NLP) tasks, such models often make unreliable predictions on out-of-distribution (OOD) inputs. As such, OOD detection is a key component of a reliable machine-learning model for any industry-scale application. Common approaches often assume access to additional OOD samples during the training stage, however, outlier distribution is often unknown in advance. Instead, we propose a post hoc framework called POORE - POsthoc pseudo-Ood REgularization, that generates pseudo-OOD samples using in-distribution (IND) data. The model is fine-tuned by introducing a new regularization loss that separates the embeddings of IND and OOD data, which leads to significant gains on the OOD prediction task during testing. We extensively evaluate our framework on three real-world dialogue systems, achieving new state-of-the-art in OOD detection.

Collaborative Anomaly Detection

In recommendation systems, items are likely to be exposed to various users and we would like to learn about the familiarity of a new user with an existing item. This can be formulated as an anomaly detection (AD) problem distinguishing between "common users" (nominal) and "fresh users" (anomalous). Considering the sheer volume of items and the sparsity of user-item paired data, independently applying conventional single-task detection methods on each item quickly becomes difficult, while correlations between items are ignored. To address this multi-task anomaly detection problem, we propose collaborative anomaly detection (CAD) to jointly learn all tasks with an embedding encoding correlations among tasks. We explore CAD with conditional density estimation and conditional likelihood ratio estimation. We found that: $i$) estimating a likelihood ratio enjoys more efficient learning and yields better results than density estimation. $ii$) It is beneficial to select a small number of tasks in advance to learn a task embedding model, and then use it to warm-start all task embeddings. Consequently, these embeddings can capture correlations between tasks and generalize to new correlated tasks.

Number Entity Recognition

Numbers are essential components of text, like any other word tokens, from which natural language processing (NLP) models are built and deployed. Though numbers are typically not accounted for distinctly in most NLP tasks, there is still an underlying amount of numeracy already exhibited by NLP models. In this work, we attempt to tap this potential of state-of-the-art NLP models and transfer their ability to boost performance in related tasks. Our proposed classification of numbers into entities helps NLP models perform well on several tasks, including a handcrafted Fill-In-The-Blank (FITB) task and on question answering using joint embeddings, outperforming the BERT and RoBERTa baseline classification.

elBERto: Self-supervised Commonsense Learning for Question Answering

Commonsense question answering requires reasoning about everyday situations and causes and effects implicit in context. Typically, existing approaches first retrieve external evidence and then perform commonsense reasoning using these evidence. In this paper, we propose a Self-supervised Bidirectional Encoder Representation Learning of Commonsense (elBERto) framework, which is compatible with off-the-shelf QA model architectures. The framework comprises five self-supervised tasks to force the model to fully exploit the additional training signals from contexts containing rich commonsense. The tasks include a novel Contrastive Relation Learning task to encourage the model to distinguish between logically contrastive contexts, a new Jigsaw Puzzle task that requires the model to infer logical chains in long contexts, and three classic SSL tasks to maintain pre-trained models language encoding ability. On the representative WIQA, CosmosQA, and ReClor datasets, elBERto outperforms all other methods, including those utilizing explicit graph reasoning and external knowledge retrieval. Moreover, elBERto achieves substantial improvements on out-of-paragraph and no-effect questions where simple lexical similarity comparison does not help, indicating that it successfully learns commonsense and is able to leverage it when given dynamic context.

Capturing Actionable Dynamics with Structured Latent Ordinary Differential Equations

End-to-end learning of dynamical systems with black-box models, such as neural ordinary differential equations (ODEs), provides a flexible framework for learning dynamics from data without prescribing a mathematical model for the dynamics. Unfortunately, this flexibility comes at the cost of understanding the dynamical system, for which ODEs are used ubiquitously. Further, experimental data are collected under various conditions (inputs), such as treatments, or grouped in some way, such as part of sub-populations. Understanding the effects of these system inputs on system outputs is crucial to have any meaningful model of a dynamical system. To that end, we propose a structured latent ODE model that explicitly captures system input variations within its latent representation. Building on a static latent variable specification, our model learns (independent) stochastic factors of variation for each input to the system, thus separating the effects of the system inputs in the latent space. This approach provides actionable modeling through the controlled generation of time-series data for novel input combinations (or perturbations). Additionally, we propose a flexible approach for quantifying uncertainties, leveraging a quantile regression formulation. Experimental results on challenging biological datasets show consistent improvements over competitive baselines in the controlled generation of observational data and prediction of biologically meaningful system inputs.