ICML 2024 Papers

In this paper, we study the feature learning ability of two-layer neural networks in the mean-field regime through the lens of kernel methods. To focus on the dynamics of the kernel induced by the first layer, we utilize a two-timescale limit, where the second layer moves much faster than the first layer. In this limit, the learning problem is reduced to the minimization problem over the intrinsic kernel. Then, we show the global convergence of the mean-field Langevin dynamics and derive time and particle discretization error. We also demonstrate that two-layer neural networks can learn a union of multiple reproducing kernel Hilbert spaces more efficiently than any kernel methods, and neural networks aquire data-dependent kernel which aligns with the target function. In addition, we develop a label noise procedure, which converges to the global optimum and show that the degrees of freedom appears as an implicit reguralization.

Unsupervised domain adaptation of multivariate time series aims to train a model to adapt its classification ability from a labeled source domain to an unlabeled target domain, where there are differences in the distribution between domains. Existing methods extract domain-invariant features directly via a shared feature extractor, neglecting the exploration of the underlying causal patterns, which undermines their reliability, especially in complex multivariate dynamic systems. To address this problem, we propose CauDiTS, an innovative framework for unsupervised domain adaptation of multivariate time series. CauDiTS adopts an adaptive rationale disentangler to disentangle domain-common causal rationales and domain-specific correlations from variable interrelationships. The stability of causal rationales across domains is vital for filtering domainspecific perturbations and facilitating the extraction of domain-invariant representations. Moreover, we promote the cross-domain consistency of intra-class causal rationales employing the learning strategies of causal prototype consistency and domain-intervention causality invariance. CauDiTS is evaluated on four benchmark datasets, demonstrating its effectiveness and outperforming state-of-the-art methods.

Sampling-based methods, e.g., Deep Ensembles and Bayesian Neural Nets have become promising approaches to improve the quality of uncertainty estimation and robust generalization. However, they suffer from a large model size and high latency at test time, which limits the scalability needed for low-resource devices and real-time applications. To resolve these computational issues, we propose Density-Softmax, a sampling-free deterministic framework via combining a density function built on a Lipschitz-constrained feature extractor with the softmax layer. Theoretically, we show that our model is the solution of minimax uncertainty risk and is distance-aware on feature space, thus reducing the over-confidence of the standard softmax under distribution shifts. Empirically, our method enjoys competitive results with state-of-the-art techniques in terms of uncertainty and robustness, while having a lower number of model parameters and a lower latency at test time.

Detecting data points deviating from the training distribution is pivotal for ensuring reliable machine learning. Extensive research has been dedicated to the challenge, spanning classical anomaly detection techniques to contemporary out-of-distribution (OOD) detection approaches. While OOD detection commonly relies on supervised learning from a labeled in-distribution (ID) dataset, anomaly detection may treat the entire ID data as a single class and disregard ID labels. This fundamental distinction raises a significant question that has yet to be rigorously explored: when and how does ID label help OOD detection? This paper bridges this gap by offering a formal understanding to theoretically delineate the impact of ID labels on OOD detection. We employ a graph-theoretic approach, rigorously analyzing the separability of ID data from OOD data in a closed-form manner. Key to our approach is the characterization of data representations through spectral decomposition on the graph. Leveraging these representations, we establish a provable error bound that compares the OOD detection performance with and without ID labels, unveiling conditions for achieving enhanced OOD detection. Lastly, we present empirical results on both simulated and real datasets, validating theoretical guarantees and reinforcing our insights.

Schrödinger bridge (SB) has emerged as the go-to method for optimizing transportation plans in diffusion models. However, SB requires estimating the intractable forward score functions, inevitably resulting in the (costly) implicit training loss based on simulated trajectories. To improve the scalability while preserving efficient transportation plans, we leverage variational inference to linearize the forward score functions (variational scores) of SB and restore *simulation-free* properties in training backward scores. We propose the variational Schrödinger diffusion model (VSDM), where the forward process is a multivariate diffusion and the variational scores are adaptively optimized for efficient transport. Theoretically, we use stochastic approximation to prove the convergence of the variational scores and show the convergence of the adaptively generated samples based on the optimal variational scores. Empirically, we test the algorithm in simulated examples and observe that VSDM is efficient in generations of anisotropic shapes and yields straighter sample trajectories compared to the single-variate diffusion. We also verify the scalability of the algorithm in real-world data and achieve competitive unconditional generation performance in CIFAR10 and conditional generation in time series modeling. Notably, VSDM no longer depends on warm-up initializations required by SB.

Motivated by recent advances in large language models for Natural Language Processing (NLP), we design a time-series foundation model for forecasting whose out-of-the-box zero-shot performance on a variety of public datasets comes close to the accuracy of state-of-the-art supervised forecasting models for each individual dataset. Our model is based on pretraining a decoder style attention model with input patching, using a large time-series corpus comprising both real-world and synthetic datasets. Experiments on a diverse set of previously unseen forecasting datasets suggests that the model can yield accurate zero-shot forecasts across different domains, forecasting horizons and temporal granularities.

Time series analysis is essential for comprehending the complexities inherent in various real-world systems and applications. Although large language models (LLMs) have recently made significant strides, the development of artificial general intelligence (AGI) equipped with time series analysis capabilities remains in its nascent phase. Most existing time series models heavily rely on domain knowledge and extensive model tuning, predominantly focusing on prediction tasks. In this paper, we argue that current LLMs have the potential to revolutionize time series analysis, thereby promoting efficient decision-making and advancing towards a more universal form of time series analytical intelligence. Such advancement could unlock a wide range of possibilities, including time series modality switching and question answering. We encourage researchers and practitioners to recognize the potential of LLMs in advancing time series analysis and emphasize the need for trust in these related efforts. Furthermore, we detail the seamless integration of time series analysis with existing LLM technologies and outline promising avenues for future research.

Time series data across scientific domains are often collected under distinct states (e.g., tasks), wherein latent processes (e.g., biological factors) create complex inter- and intra-state variability. A key approach to capture this complexity is to uncover fundamental interpretable units within the data, Building Blocks (BBs), which modulate their activity and adjust their structure across observations. Existing methods for identifying BBs in multi-way data often overlook inter- vs. intra-state variability, produce uninterpretable components, or do not align with properties of real-world data, such as missing samples and sessions of different duration. Here, we present a framework for Similarity-driven Building Block Inference using Graphs across States (SiBBlInGS). SiBBlInGS offers a graph-based dictionary learning approach for discovering sparse BBs along with their temporal traces, based on co-activity patterns and inter- vs. intra-state relationships. Moreover, SiBBlInGS captures per-trial temporal variability and controlled cross-state structural BB adaptations, identifies state-specific vs. state-invariant components, and accommodates variability in the number and duration of observed sessions across states. We demonstrate SiBBlInGS's ability to reveal insights into complex phenomena as well as its robustness to noise and missing samples through several synthetic and real-world examples, including web search and neural data.

We study the infinite-horizon distributionally robust (DR) control of linear systems with quadratic costs, where disturbances have unknown, possibly time-correlated distribution within a Wasserstein-2 ambiguity set. We aim to minimize the worst-case expected regret—the excess cost of a causal policy compared to a non-causal one with access to future disturbance. Though the optimal policy lacks a finite-order state-space realization (i.e., it is non-rational), it can be characterized by a finite-dimensional parameter. Leveraging this, we develop an efficient frequency-domain algorithm to compute this optimal control policy and present a convex optimization method to construct a near-optimal state-space controller that approximates the optimal non-rational controller in the $\mathit{H}_\infty$-norm. This approach avoids solving a computationally expensive semi-definite program (SDP) that scales with the time horizon in the finite-horizon setting.

Learning Continuous-Time Dynamic Graphs (C-TDGs) requires accurately modeling spatio-temporal information on streams of irregularly sampled events. While many methods have been proposed recently, we find that most message passing-, recurrent- or self-attention-based methods perform poorly on *long-range* tasks. These tasks require correlating information that occurred "far" away from the current event, either spatially (higher-order node information) or along the time dimension (events occurred in the past). To address long-range dependencies, we introduce Continuous-Time Graph Anti-Symmetric Network (CTAN). Grounded within the ordinary differential equations framework, our method is designed for efficient propagation of information. In this paper, we show how CTAN's (i) long-range modeling capabilities are substantiated by theoretical findings and how (ii) its empirical performance on synthetic long-range benchmarks and real-world benchmarks is superior to other methods. Our results motivate CTAN's ability to propagate long-range information in C-TDGs as well as the inclusion of long-range tasks as part of temporal graph models evaluation.

Learning a decent representation from unlabeled time series is a challenging task, especially when the time series data is derived from diverse channels at different sampling rates. Our motivation stems from the financial domain, where sparsely labeled covariates are commonly collected at different frequencies, *e.g.*, daily stock market index, monthly unemployment rate and quarterly net revenue of a certain listed corporation. This paper presents **M**ulti-**F**requency **C**ontrastive **L**earning **R**epresentation (MF-CLR), aimed at learning a good representation of multi-frequency time series in a self-supervised paradigm by leveraging the ability of contrastive learning. MF-CLR introduces a hierarchical mechanism that spans across different frequencies along the feature dimension. Within each contrastive block, two groups of subseries with adjacent frequencies are embedded based on our proposed cross-frequency consistency. To validate the effectiveness of MF-CLR, we conduct extensive experiments on five downstream tasks, including long-term and short-term forecasting, classification, anomaly detection and imputation. Experimental evidence shows that MF-CLR delivers a leading performance in all the downstream tasks and keeps consistent performance across different target dataset scales in the transfer learning scenario.

Temporal difference (TD) learning algorithms with neural network function parameterization have well-established empirical success in many practical large-scale reinforcement learning tasks. However, theoretical understanding of these algorithms remains challenging due to the nonlinearity of the action-value approximation. In this paper, we develop an improved non-asymptotic analysis of the neural TD method with a general $L$-layer neural network. New proof techniques are developed and an improved new $\tilde{\mathcal{O}}(\epsilon^{-1})$ sample complexity is derived. To our best knowledge, this is the first finite-time analysis of neural TD that achieves an $\tilde{\mathcal{O}}(\epsilon^{-1})$ complexity under the Markovian sampling, as opposed to the best known $\tilde{\mathcal{O}}(\epsilon^{-2})$ complexity in the existing literature.

This paper introduces the Deep Functional Factor Model (DF2M), a Bayesian nonparametric model designed for analysis of high-dimensional functional time series. DF2M is built upon the Indian Buffet Process and the multi-task Gaussian Process, incorporating a deep kernel function that captures non-Markovian and nonlinear temporal dynamics. Unlike many black-box deep learning models, DF2M offers an explainable approach to utilizing neural networks by constructing a factor model and integrating deep neural networks within the kernel function. Additionally, we develop a computationally efficient variational inference algorithm to infer DF2M. Empirical results from four real-world datasets demonstrate that DF2M provides better explainability and superior predictive accuracy compared to conventional deep learning models for high-dimensional functional time series.

We study the problem of maintaining a decision tree in the fully-dynamic setting, where the dataset is updated by an adversarial sequence of insertions and deletions. We present the first algorithm with strong guarantees on both the quality of the tree and the worst-case update time (the maximum time spent between two consecutive dataset updates). For instance, we can maintain a tree where each node has Gini gain within $\beta$ of the optimum, while guaranteeing an update time $O(d \beta^{-3} \log^4 n )$, where $d$ is the number of features and $n$ the maximum size of the dataset. This is optimal up to polylogarithmic factors, as any dynamic algorithm must have update time in $\Omega(d)$. Similar guarantees hold for the variance and information gain, for classification and regression, and even for *boosted* trees. This shows that many popular decision trees such as ID3 or C4.5 can be efficiently be made dynamic, answering an open question of Bressan, Damay and Sozio (AAAI 2023). We also show that, under the 3SUM conjecture or the Orthogonal Vectors Hypothesis, the update time must be polynomial in $1/\beta$.

Anomaly detection is vital in many domains, such as finance, healthcare, and cybersecurity. In this paper, we propose a novel deep anomaly detection method for tabular data that leverages Non-Parametric Transformers (NPTs), a model initially proposed for supervised tasks, to capture both feature-feature and sample-sample dependencies. In a reconstruction-based framework, we train an NPT to reconstruct masked features of normal samples. In a non-parametric fashion, we leverage the whole training set during inference and use the model's ability to reconstruct the masked features to generate an anomaly score. To the best of our knowledge, this is the first work to successfully combine feature-feature and sample-sample dependencies for anomaly detection on tabular datasets. Through extensive experiments on 31 benchmark tabular datasets, we demonstrate that our method achieves state-of-the-art performance, outperforming existing methods by 2.4% and 1.2% in terms of F1-score and AUROC, respectively. Our ablation study further proves that modeling both types of dependencies is crucial for anomaly detection on tabular data.

In real-world applications, time series data frequently exhibit non-stationarity, with statistics changing over time. This variability undermines the forecasting accuracy of deep learning models that are trained on historical data but deployed for future prediction. A common approach to mitigate this issue involves normalizing the data to counteract statistical drift, followed by denormalization on the prediction. However, existing methods often employ heuristic normalization techniques that do not fully account for the unique characteristics of the series. Our paper addresses the critical question in this context: which statistics should be removed and restored? We argue that the statistics selected for normalization should exhibit both local invariance and global variability to ensure their correctness and helpfulness. To this end, we propose the Selective and Interpretable Normalization methodology, dubbed SIN. This approach maximizes the covariance between a given look-back window and its subsequent future values, thereby identifying key statistics for normalization and simultaneously learning the corresponding normalization transformations. The interpretable framework can be used to explain the success and limitations of some popular normalization methods. By integrating SIN, we demonstrate improvements in the performance of several prevalent forecasting models, thereby validating the utility of our approach.

Discovering causal relationships from time series data is significant in fields such as finance, climate science, and neuroscience. However, contemporary techniques rely on the simplifying assumption that data originates from the same causal model, while in practice, data is heterogeneous and can stem from different causal models. In this work, we relax this assumption and perform causal discovery from time series data originating from *a mixture of causal models*. We propose a general variational inference-based framework called MCD to infer the underlying causal models as well as the mixing probability of each sample. Our approach employs an end-to-end training process that maximizes an evidence-lower bound for the data likelihood. We present two variants: MCD-Linear for linear relationships and independent noise, and MCD-Nonlinear for nonlinear causal relationships and history-dependent noise. We demonstrate that our method surpasses state-of-the-art benchmarks in causal discovery tasks through extensive experimentation on synthetic and real-world datasets, particularly when the data emanates from diverse underlying causal graphs. Theoretically, we prove the identifiability of such a model under some mild assumptions. Implementation is available at [https://github.com/Rose-STL-Lab/MCD](https://github.com/Rose-STL-Lab/MCD).

Concept guidance has emerged as a cheap and simple way to control the behavior of language models by probing their hidden representations for concept vectors and using them to perturb activations at inference time. While the focus of previous work has largely been on *truthfulness*, in this paper we extend this framework to a richer set of concepts such as *appropriateness*, *humor*, *creativity* and *quality*, and explore to what degree current detection and guidance strategies work in these challenging settings. To facilitate evaluation, we develop a novel metric for concept guidance that takes into account both the success of concept elicitation as well as the potential degradation in fluency of the guided model. Our extensive experiments reveal that while some concepts such as *truthfulness* more easily allow for guidance with current techniques, novel concepts such as *appropriateness* or *humor* either remain difficult to elicit, need extensive tuning to work, or even experience confusion. Moreover, we find that probes with optimal detection accuracies do not necessarily make for the optimal guides, contradicting previous observations for *truthfulness*. Our work warrants a deeper investigation into the interplay between detectability, guidability, and the nature of the concept, and we hope that our rich experimental test-bed for guidance research inspires stronger follow-up approaches.

Language models deployed in the wild make errors. However, simply updating the model with the corrected error instances causes catastrophic forgetting---the updated model makes errors on instances learned during the instruction tuning or upstream training phase. Randomly replaying upstream data yields unsatisfactory performance and often comes with high variance and poor controllability. To this end, we try to forecast upstream examples that will be forgotten due to a model update for improved controllability of the replay process and interpretability. We train forecasting models given a collection of online learned examples and corresponding forgotten upstream pre-training examples. We propose a partially interpretable forecasting model based on the observation that changes in pre-softmax logit scores of pretraining examples resemble that of online learned examples, which performs decently on BART but fails on T5 models. We further show a black-box classifier based on inner products of example representations achieves better forecasting performance over a series of setups. Finally, we show that we reduce forgetting of upstream pretraining examples by replaying examples that are forecasted to be forgotten, demonstrating the practical utility of forecasting example forgetting.