UAI 2023 Papers

Unsupervised anomaly detection is a binary classification that detects anomalies in unseen samples given only unlabeled normal data. Reconstruction-based approaches are widely used, which perform reconstruction error minimization on training data to learn normal patterns and quantify the degree of anomalies by reconstruction errors on testing data. However, this approach tends to miss anomalies when the normal data has multi-pattern. Because the model generalizes unrestrictedly beyond normal patterns even to include anomaly patterns. In this paper, we proposed a memory mechanism that memorizes typical normal patterns through a capacity-controlled external differentiable matrix so that the generalization of the model to anomalies is limited by the retrieval of the matrix. We achieved state-of-the-art performance on several public benchmarks.

We consider the problem of maximizing submodular functions under submodular constraints by formulating the problem in two ways: SCSKC and DiffC. Given two submodular functions f and g where f is monotone, the objective of SCSKC problem is to find a set S of size at most k that maximizes f(S) under the constraint that g(S) < theta, for a given value of theta. The problem of DiffC focuses on finding a set S of size at most k such that h(S) = f(S)-g(S) is maximized. It is known that these problems are highly inapproximable and do not admit any constant factor multiplicative approximation algorithms unless NP is easy. Known approximation algorithms involve data-dependent approximation factors that are not efficiently computable. We initiate a study of the design of approximation algorithms where the approximation factors are efficiently computable. For the problem of SCSKC, we prove that the greedy algorithm produces a solution whose value is at least (1-1/e)f(OPT) - A, where A is the data-dependent additive error. For the DiffC problem, we design an algorithm that uses the SCSKC greedy algorithm as a subroutine. This algorithm produces a solution whose value is at least (1-1/e)h(OPT)-B, where B is also a data-dependent additive error. A salient feature of our approach is that the additive error terms can be computed efficiently, thus enabling us to ascertain the quality of the solutions produced.

Reweighted wake-sleep (RWS) is a machine learning method for performing Bayesian inference in a very general class of models. RWS draws $K$ samples from an underlying approximate posterior, then uses importance weighting to provide a better estimate of the true posterior. RWS then updates its approximate posterior towards the importance-weighted estimate of the true posterior. However, recent work [Chattergee and Diaconis, 2018] indicates that the number of samples required for effective importance weighting is exponential in the number of latent variables. Attaining such a large number of importance samples is intractable in all but the smallest models. Here, we develop massively parallel RWS, which circumvents this issue by drawing $K$ samples of all $n$ latent variables, and individually reasoning about all $K^n$ possible combinations of samples. While reasoning about $K^n$ combinations might seem intractable, the required computations can be performed in polynomial time by exploiting conditional independencies in the generative model. We show considerable improvements over standard “global” RWS, which draws $K$ samples from the full joint.

Entity linking, bridging mentions in the contexts with their corresponding entities in the knowledge bases, has attracted wide attention due to many potential applications. Recently, plenty of multimodal entity linking approaches have been proposed to take full advantage of the visual information rather than solely the textual modality. Although feasible, these methods mainly focus on the single-mention scenarios and neglect the scenarios where multiple mentions exist simultaneously in the same context, which limits the performance. In fact, such multi-mention scenarios are pretty common in public datasets and real-world applications. To solve this challenge, we first propose a joint feature extraction module to learn the representations of context and entity candidates, from both the visual and textual perspectives. Then, we design a pairwise training scheme (for training) and a multi-mention collaborative ranking method (for testing) to model the potential connections between different mentions. We evaluate our method on a public dataset and a self-constructed dataset, NYTimes-MEL, under both text-only and multimodal scenarios. The experimental results demonstrate that our method can largely outperform the state-of-the-art methods, especially in multi-mention scenarios. Our dataset and source code are publicly available at https://github.com/ycm094/MMEL-main.

Physical adversarial attacks can mislead detectors in real-world scenarios and have attracted increasing attention. However, most existing works manipulate the detector’s final outputs as attack targets while ignoring the inherent characteristics of objects. This can result in attacks being trapped in model-specific local optima and reduced transferability. To address this issue, we propose a

One of the major challenges in multi-person pose estimation is instance-aware keypoint estimation. Previous methods address this problem by leveraging an off-the-shelf detector, heuristic post-grouping process or explicit instance identification process, hindering further improvements in the inference speed which is an important factor for practical applications. From the statistical point of view, those additional processes for identifying instances are necessary to bypass learning the high-dimensional joint distribution of human keypoints, which is a critical factor for another major challenge, the occlusion scenario. In this work, we propose a novel framework of single-stage instance-aware pose estimation by modeling the joint distribution of human keypoints with a mixture density model, termed as MDPose. Our MDPose estimates the distribution of human keypoints’ coordinates using a mixture density model with an instance-aware keypoint head consisting simply of 8 convolutional layers. It is trained by minimizing the negative log-likelihood of the ground truth keypoints. Also, we propose a simple yet effective training strategy, Random Keypoint Grouping (RKG), which significantly alleviates the underflow problem leading to successful learning of relations between keypoints. On OCHuman dataset, which consists of images with highly occluded people, our MDPose achieves state-of-the-art performance by successfully learning the high-dimensional joint distribution of human keypoints. Furthermore, our MDPose shows significant improvement in inference speed with a competitive accuracy on MS COCO, a widely-used human keypoint dataset, thanks to the proposed much simpler single-stage pipeline.

We study a matrix recovery problem with unknown correspondence: given the observation matrix $M_o=[A,\tilde P B]$, where $\tilde P$ is an unknown permutation matrix, we aim to recover the underlying matrix $M=[A,B]$. Such problem commonly arises in many applications where heterogeneous data are utilized and the correspondence among them are unknown, e.g., due to data mishandling or privacy concern. We show that, in some applications, it is possible to recover $M$ via solving a nuclear norm minimization problem. Moreover, under a proper low-rank condition on $M$, we derive a non-asymptotic error bound for the recovery of $M$. We propose an algorithm, $\text{M}^3\text{O}$ (Matrix recovery via Min-Max Optimization) which recasts this combinatorial problem as a continuous minimax optimization problem and solves it by proximal gradient with a Max-Oracle. $\text{M}^3\text{O}$ can also be applied to a more general scenario where we have missing entries in $M_o$ and multiple groups of data with distinct unknown correspondence. Experiments on simulated data, the MovieLens 100K dataset and Yale B database show that $\text{M}^3\text{O}$ achieves state-of-the-art performance over several baselines and can recover the ground-truth correspondence with high accuracy.

There has been significant interest in searching for off-policy Temporal-Difference (TD) algorithms that find the same solution that would have been obtained in the on-policy regime. An important property of such algorithms is that their expected update has the same fixed point as that of On-policy TD($\lambda$), which we call

Efficiently and reliably estimating uncertainty is an important objective in deep learning. It is especially pertinent to autoregressive sequence tasks, where training and inference costs are typically very high. However, existing research has predominantly focused on tasks with static data such as image classification. In this work, we investigate Ensemble Distribution Distillation (EDD) applied to large-scale natural language sequence-to-sequence data. EDD aims to compress the superior uncertainty performance of an expensive (teacher) ensemble into a cheaper (student) single model. Importantly, the ability to separate knowledge (epistemic) and data (aleatoric) uncertainty is retained. Existing probability-space approaches to EDD, however, are difficult to scale to large vocabularies. We show, for modern transformer architectures on large-scale translation tasks, that modelling the ensemble

Sparse graphs built by sparse representation has been demonstrated to be effective in clustering high-dimensional data. Albeit the compelling empirical performance, the vanilla sparse graph ignores the geometric information of the data by performing sparse representation for each datum separately. In order to obtain a sparse graph aligned with the local geometric structure of data, we propose a novel Support Regularized Sparse Graph, abbreviated as SRSG, for data clustering. SRSG encourages local smoothness on the neighborhoods of nearby data points by a well-defined support regularization term. We propose a fast proximal gradient descent method to solve the non-convex optimization problem of SRSG with the convergence matching the Nesterov’s optimal convergence rate of first-order methods on smooth and convex objective function with Lipschitz continuous gradient. Extensive experimental results on various real data sets demonstrate the superiority of SRSG over other competing clustering methods.

Message passing on factor graphs is a powerful framework for probabilistic inference, which finds important applications in various scientific domains. The most wide-spread message passing scheme is the sum-product algorithm (SPA) which gives exact results on trees but often fails on graphs with many small cycles. We search for an alternative message passing algorithm that works particularly well on such cyclic graphs. Therefore, we challenge the extrinsic principle of the SPA, which loses its objective on graphs with cycles. We further replace the local SPA message update rule at the factor nodes of the underlying graph with a generic mapping, which is optimized in a data-driven fashion. These modifications lead to a considerable improvement in performance while preserving the simplicity of the SPA. We evaluate our method for two classes of cyclic graphs: the 2x2 fully connected Ising grid and factor graphs for symbol detection on linear communication channels with inter-symbol interference. To enable the method for large graphs as they occur in practical applications, we develop a novel loss function that is inspired by the Bethe approximation from statistical physics and allows for training in an unsupervised fashion.

Machine learning algorithms are often repeatedly. applied to problems with similar structure over and over again. We focus on solving a sequence of bandit optimization tasks and develop LIBO, an algorithm which adapts to the environment by learning from past experience and becomes more sample-efficient in the process. We assume a kernelized structure where the kernel is unknown but shared across all tasks. LIBO sequentially meta-learns a kernel that approximates the true kernel and solves the incoming tasks with the latest kernel estimate. Our algorithm can be paired with any kernelized or linear bandit algorithm and guarantees oracle optimal performance, meaning that as more tasks are solved, the regret of LIBO on each task converges to the regret of the bandit algorithm with oracle knowledge of the true kernel. Naturally, if paired with a sublinear bandit algorithm, LIBO yields a sublinear lifelong regret. We also show that direct access to the data from each task is not necessary for attaining sublinear regret. We propose F-LIBO, which solves the lifelong problem in a federated manner.

Sequential decision-making (SDM) methods enable AI agents to compute an action policy toward achieving long-term goals under uncertainty. Existing research has shown that contextual knowledge in declarative forms can be used for improving the performance of SDM methods. However, the contextual knowledge from people tends to be incomplete and sometimes inaccurate, which greatly limits the applicability of knowledge-based SDM methods. In this paper, we develop a novel algorithm for knowledge-based SDM, called PERIL, that learns from interaction experience to reason about contextual knowledge, as applied to urban driving scenarios. Experiments have been conducted using CARLA, a widely used autonomous driving simulator. Results demonstrate PERIL’s superiority in comparison to existing knowledge-based SDM baselines.

Reinforcement learning algorithms have achieved remarkable success in acquiring behavioral skills directly from pixel inputs. However, their application in real-world scenarios presents challenges due to their sensitivity to visual distractions (e.g., changes in viewpoint and light). A key factor contributing to this challenge is that the learned representations often suffer from overfitting task-irrelevant information. By comparing several representation learning methods, we find that the key to alleviating overfitting in representation learning is to choose proper prediction targets. Motivated by our comparison, we propose a novel representation learning approach—namely, reward sequence prediction (RSP)—that uses reward sequences or their transforms (e.g., discrete time Fourier transform) as prediction targets. RSP can efficiently learn robust representations as reward sequences rarely contain task-irrelevant information while providing a large number of supervised signals to accelerate representation learning. An appealing feature is that RSP makes no assumption about the type of distractions and thus can improve performance even when multiple types of distractions exist. We evaluate our approach in Distracting Control Suite. Experiments show that our method achieves state-of-the-art sample efficiency and generalization ability in tasks with distractions.

It is a remarkable fact that the same $O(\sqrt{T})$ regret rate can be achieved in both the Experts Problem and the Adversarial Multi-Armed Bandit problem albeit with a worse dependence on number of actions in the latter case. In contrast, it has been shown that handling online MDPs with communicating structure and bandit information incurs $\Omega(T^{2/3})$ regret even in the case of deterministic transitions. Is this the price we pay for handling communicating structure or is it because we also have bandit feedback? In this paper we show that with full information, online MDPs can still be learned at an $O(\sqrt{T})$ rate even in the presence of communicating structure. We first show this by proposing an efficient follow the perturbed leader (FPL) algorithm for the deterministic transition case. We then extend our scope to consider stochastic transitions where we first give an inefficient $O(\sqrt{T})$-regret algorithm (with a mild additional condition on the dynamics). Then we show how to achieve $O\left(\sqrt{\frac{T}{\alpha}}\right)$ regret rate using an oracle-efficient algorithm but with the additional restriction that the starting state distribution has mass at least $\alpha$ on each state.

We study the causal bandit problem that entails identifying a near-optimal intervention from a specified set A of (possibly non-atomic) interventions over a given causal graph. Here, an optimal intervention in A is one that maximizes the expected value for a designated reward variable in the graph, and we use the standard notion of simple regret to quantify near optimality. Considering Bernoulli random variables and for causal graphs on N vertices with constant in-degree, prior work has achieved a worst case guarantee of O(N/sqrt(T)) for simple regret. The current work utilizes the idea of covering interventions (which are not necessarily contained within A) and establishes a simple regret guarantee of O(sqrt(N/T)). Notably, and in contrast to prior work, our simple regret bound depends only on explicit parameters of the problem instance. We also go beyond prior work and achieve a simple regret guarantee for causal graphs with unobserved variables. Further, we perform experiments to show improvements over baselines in this setting.

Under a simplified data model, this paper provides a theoretical analysis of learning from data that have an underlying low-rank tensor structure in both supervised and unsupervised settings. For the supervised setting, we provide an analysis of a Ridge classifier (with high regularization parameter) with and without knowledge of the low-rank structure of the data. Our results quantify analytically the gain in misclassification errors achieved by exploiting the low-rank structure for denoising purposes, as opposed to treating data as mere vectors. We further provide a similar analysis in the context of clustering, thereby quantifying the exact performance gap between tensor methods and standard approaches which treat data as simple vectors.

Gradient inversion attack enables recovery of training samples from model gradients in federated learning (FL), and constitutes a serious threat to data privacy. To mitigate this vulnerability, prior work proposed both principled defenses based on differential privacy, as well as heuristic defenses based on gradient compression as countermeasures. These defenses have so far been very effective, in particular those based on gradient compression that allow the model to maintain high accuracy while greatly reducing the effectiveness of attacks. In this work, we argue that such findings underestimate the privacy risk in FL. As a counterexample, we show that existing defenses can be broken by a simple adaptive attack, where a model trained on auxiliary data is able to invert gradients on both vision and language tasks.

Learning causal effects is a fundamental problem in science. Anchor regression has been developed to address this problem for a large class of causal graphical models, though the relationships between the variables are assumed to be linear. In this work, we tackle the nonlinear setting by proposing kernel anchor regression (KAR). Beyond a classic two-stage least square (2SLS) estimator, we also study an improved variant that involves nonparametric kernel regression in three separate stages. We provide convergence results for the proposed KAR estimators and the identifiability conditions for KAR to learn the nonlinear structural equation models (SEM). Experimental results demonstrate the superior performances of the proposed KAR estimators over existing baselines.