Chenghao Liu

Stephan Xie, Ben Cohen, Mononito Goswami et al. · cmu

Time series question-answering (TSQA), in which we ask natural language questions to infer and reason about properties of time series, is a promising yet underexplored capability of foundation models. In this work, we present ARFBench, a TSQA benchmark that evaluates the understanding of multimodal foundation models (FMs) on time series anomalies prevalent in software incident data. ARFBench consists of 750 questions across 142 time series and 5.38M data points from 63 production incidents sourced exclusively from internal telemetry at Datadog. We evaluate leading proprietary and open-source LLMs, VLMs, and time series FMs and observe that frontier VLMs perform markedly better than existing baselines; the leading model (GPT-5) achieves a 62.7% accuracy and 51.9% F1. We next demonstrate the promise of specialized multimodal approaches. We develop a novel TSFM + VLM hybrid prototype which we post-train on a small set of synthetic and real data that yields comparable overall F1 and accuracy with frontier models. Lastly, we find models and human domain experts exhibit complementary strengths. We define a model-expert oracle, a best-of-2 oracle selector over model and expert answers, yielding 82.8% F1 and 87.2% accuracy and establishing a new superhuman frontier for future TSQA models. The benchmark is available at https://huggingface.co/datasets/Datadog/ARFBench.

Devansh Arpit, Matthew Fernandez, Itai Feigenbaum et al. · salesforce, stanford

We introduce the Salesforce CausalAI Library, an open-source library for causal analysis using observational data. It supports causal discovery and causal inference for tabular and time series data, of discrete, continuous and heterogeneous types. This library includes algorithms that handle linear and non-linear causal relationships between variables, and uses multi-processing for speed-up. We also include a data generator capable of generating synthetic data with specified structural equation model for the aforementioned data formats and types, that helps users control the ground-truth causal process while investigating various algorithms. Finally, we provide a user interface (UI) that allows users to perform causal analysis on data without coding. The goal of this library is to provide a fast and flexible solution for a variety of problems in the domain of causality. This technical report describes the Salesforce CausalAI API along with its capabilities, the implementations of the supported algorithms, and experiments demonstrating their performance and speed. Our library is available at \url{https://github.com/salesforce/causalai}.

Quang Pham, Chenghao Liu, Steven C. H. Hoi

According to the Complementary Learning Systems (CLS) theory~\cite{mcclelland1995there} in neuroscience, humans do effective \emph{continual learning} through two complementary systems: a fast learning system centered on the hippocampus for rapid learning of the specifics, individual experiences; and a slow learning system located in the neocortex for the gradual acquisition of structured knowledge about the environment. Motivated by this theory, we propose \emph{DualNets} (for Dual Networks), a general continual learning framework comprising a fast learning system for supervised learning of pattern-separated representation from specific tasks and a slow learning system for representation learning of task-agnostic general representation via Self-Supervised Learning (SSL). DualNets can seamlessly incorporate both representation types into a holistic framework to facilitate better continual learning in deep neural networks. Via extensive experiments, we demonstrate the promising results of DualNets on a wide range of continual learning protocols, ranging from the standard offline, task-aware setting to the challenging online, task-free scenario. Notably, on the CTrL~\cite{veniat2020efficient} benchmark that has unrelated tasks with vastly different visual images, DualNets can achieve competitive performance with existing state-of-the-art dynamic architecture strategies~\cite{ostapenko2021continual}. Furthermore, we conduct comprehensive ablation studies to validate DualNets efficacy, robustness, and scalability. Code will be made available at \url{https://github.com/phquang/DualNet}.

Gerald Woo, Chenghao Liu, Doyen Sahoo et al.

Deep learning has been actively applied to time series forecasting, leading to a deluge of new methods, belonging to the class of historical-value models. Yet, despite the attractive properties of time-index models, such as being able to model the continuous nature of underlying time series dynamics, little attention has been given to them. Indeed, while naive deep time-index models are far more expressive than the manually predefined function representations of classical time-index models, they are inadequate for forecasting, being unable to generalize to unseen time steps due to the lack of inductive bias. In this paper, we propose DeepTime, a meta-optimization framework to learn deep time-index models which overcome these limitations, yielding an efficient and accurate forecasting model. Extensive experiments on real world datasets in the long sequence time-series forecasting setting demonstrate that our approach achieves competitive results with state-of-the-art methods, and is highly efficient. Code is available at https://github.com/salesforce/DeepTime.

Chenghao Liu, Wenzhuo Yang, Himanshu Mittal et al.

We introduce PyRCA, an open-source Python machine learning library of Root Cause Analysis (RCA) for Artificial Intelligence for IT Operations (AIOps). It provides a holistic framework to uncover the complicated metric causal dependencies and automatically locate root causes of incidents. It offers a unified interface for multiple commonly used RCA models, encompassing both graph construction and scoring tasks. This library aims to provide IT operations staff, data scientists, and researchers a one-step solution to rapid model development, model evaluation and deployment to online applications. In particular, our library includes various causal discovery methods to support causal graph construction, and multiple types of root cause scoring methods inspired by Bayesian analysis, graph analysis and causal analysis, etc. Our GUI dashboard offers practitioners an intuitive point-and-click interface, empowering them to easily inject expert knowledge through human interaction. With the ability to visualize causal graphs and the root cause of incidents, practitioners can quickly gain insights and improve their workflow efficiency. This technical report introduces PyRCA's architecture and major functionalities, while also presenting benchmark performance numbers in comparison to various baseline models. Additionally, we demonstrate PyRCA's capabilities through several example use cases.

Gerald Woo, Chenghao Liu, Akshat Kumar et al.

Time series has been left behind in the era of pre-training and transfer learning. While research in the fields of natural language processing and computer vision are enjoying progressively larger datasets to train massive models, the most popular time series datasets consist of only tens of thousands of time steps, limiting our ability to study the effectiveness of pre-training and scaling. Recent studies have also cast doubt on the need for expressive models and scale. To alleviate these issues, we introduce three large-scale time series forecasting datasets from the cloud operations (CloudOps) domain, the largest having billions of observations, enabling further study into pre-training and scaling of time series models. We build the empirical groundwork for studying pre-training and scaling of time series models and pave the way for future research by identifying a promising candidate architecture. We show that it is a strong zero-shot baseline and benefits from further scaling, both in model and dataset size. Accompanying these datasets and results is a suite of comprehensive benchmark results comparing classical and deep learning baselines to our pre-trained method - achieving a 27% reduction in error on the largest dataset. Code and datasets can be found https://github.com/SalesforceAIResearch/pretrain-time-series-cloudops.

Junnan Nie, Jiayi Li, Jiachen Zhang et al.

Recent vision-language-action and diffusion-based robot policies often use action chunking, where each policy query predicts a sequence of future actions and the robot executes an open-loop prefix before re-querying. While this interface improves local motion continuity, deployment still requires choosing the execution horizon: how much of each predicted chunk should be executed before acquiring a new observation. However, our experiments show that success is strongly task-dependent and non-monotonic with respect to the execution horizon, making a single constant horizon an unreliable deployment rule. We propose PACE (Phase-Aware Chunk Execution), a training-free test-time execution method that selects the execution horizon online from the predicted chunk itself. PACE exploits the phase-dependent kinematic structure of manipulation trajectories by identifying low-speed transition points in the predicted speed profile and using them as candidate replanning boundaries. Because PACE uses only the predicted action chunk, it is plug-and-play and requires no retraining or access to policy internals. We validate PACE through large-scale evaluations in both simulation and real-robot settings. On 50 RoboTwin2.0 tasks, PACE raises the average success rate from 57.8% to 64.2%. In real-robot experiments on bimanual ALOHA and single-arm Franka platforms, PACE improves the average task score from 60.7 to 77.7 and the average success rate from 50.7% to 70.4%. Ablations and rollout-level analyses show that PACE adapts execution horizons across manipulation phases, shortening near transitions while preserving longer execution during coherent motion.

Qian Cheng, Amrita Saha, Wenzhuo Yang et al.

Software and System logs record runtime information about processes executing within a system. These logs have become the most critical and ubiquitous forms of observability data that help developers understand system behavior, monitor system health and resolve issues. However, the volume of logs generated can be humongous (of the order of petabytes per day) especially for complex distributed systems, such as cloud, search engine, social media, etc. This has propelled a lot of research on developing AI-based log based analytics and intelligence solutions that can process huge volume of raw logs and generate insights. In order to enable users to perform multiple types of AI-based log analysis tasks in a uniform manner, we introduce LogAI (https://github.com/salesforce/logai), a one-stop open source library for log analytics and intelligence. LogAI supports tasks such as log summarization, log clustering and log anomaly detection. It adopts the OpenTelemetry data model, to enable compatibility with different log management platforms. LogAI provides a unified model interface and provides popular time-series, statistical learning and deep learning models. Alongside this, LogAI also provides an out-of-the-box GUI for users to conduct interactive analysis. With LogAI, we can also easily benchmark popular deep learning algorithms for log anomaly detection without putting in redundant effort to process the logs. We have opensourced LogAI to cater to a wide range of applications benefiting both academic research and industrial prototyping.

Qian Cheng, Doyen Sahoo, Amrita Saha et al.

Artificial Intelligence for IT operations (AIOps) aims to combine the power of AI with the big data generated by IT Operations processes, particularly in cloud infrastructures, to provide actionable insights with the primary goal of maximizing availability. There are a wide variety of problems to address, and multiple use-cases, where AI capabilities can be leveraged to enhance operational efficiency. Here we provide a review of the AIOps vision, trends challenges and opportunities, specifically focusing on the underlying AI techniques. We discuss in depth the key types of data emitted by IT Operations activities, the scale and challenges in analyzing them, and where they can be helpful. We categorize the key AIOps tasks as - incident detection, failure prediction, root cause analysis and automated actions. We discuss the problem formulation for each task, and then present a taxonomy of techniques to solve these problems. We also identify relatively under explored topics, especially those that could significantly benefit from advances in AI literature. We also provide insights into the trends in this field, and what are the key investment opportunities.

Mouxiang Chen, Lefei Shen, Zhuo Li et al.

Foundation models have emerged as a promising approach in time series forecasting (TSF). Existing approaches either repurpose large language models (LLMs) or build large-scale time series datasets to develop TSF foundation models for universal forecasting. However, these methods face challenges due to the severe cross-domain gap or in-domain heterogeneity. This paper explores a new road to building a TSF foundation model from rich, high-quality natural images. Our key insight is that a visual masked autoencoder, pre-trained on the ImageNet dataset, can naturally be a numeric series forecaster. By reformulating TSF as an image reconstruction task, we bridge the gap between image pre-training and TSF downstream tasks. Surprisingly, without further adaptation in the time series domain, the proposed VisionTS could achieve better zero-shot forecast performance than existing TSF foundation models. With fine-tuning for one epoch, VisionTS could further improve the forecasting and achieve state-of-the-art performance in most cases. Extensive experiments reveal intrinsic similarities between images and real-world time series, suggesting that visual models may offer a "free lunch" for TSF and highlight the potential for future cross-modality research. Our code is publicly available at https://github.com/Keytoyze/VisionTS.

Zhongzheng Qiao, Sheng Pan, Anni Wang et al.

Time series foundation models (TSFMs) are revolutionizing the forecasting landscape from specific dataset modeling to generalizable task evaluation. However, we contend that existing benchmarks exhibit common limitations in four dimensions: constrained data composition dominated by reused legacy sources, compromised data integrity lacking rigorous quality assurance, misaligned task formulations detached from real-world contexts, and rigid analysis perspectives that obscure generalizable insights. To bridge these gaps, we introduce TIME, a next-generation task-centric benchmark comprising 50 fresh datasets and 98 forecasting tasks, tailored for strict zero-shot TSFM evaluation free from data leakage. Integrating large language models and human expertise, we establish a rigorous human-in-the-loop benchmark construction pipeline to ensure high data integrity and redefine task formulation by aligning forecasting configurations with real-world operational requirements and variate predictability. Furthermore, we propose a novel pattern-level evaluation perspective that moves beyond traditional dataset-level evaluations based on static meta labels. By leveraging structural time series features to characterize intrinsic temporal properties, this approach offers generalizable insights into model capabilities across diverse patterns. We evaluate 12 representative TSFMs and establish a multi-granular leaderboard to facilitate in-depth analysis and visualized inspection. The leaderboard is available at https://huggingface.co/spaces/Real-TSF/TIME-leaderboard.

Chen Yang, Chufan Yu, Hanfu Chen et al.

MOSS-Audio is a unified audio-language model for speech, environmental sound, and music understanding, supporting audio captioning, time-aware question answering, timestamped transcription, and audio-grounded reasoning. MOSS-Audio couples a dedicated audio encoder with a modality adapter and a large language model: the encoder produces 12.5 Hz temporal representations, the adapter projects them into the decoder space, and the decoder generates autoregressive text outputs. Two design choices are central to the system: \textbf{DeepStack cross-layer feature injection}, which exposes the decoder to acoustic information from multiple encoder depths, and \textbf{time markers}, which provide explicit temporal cues by inserting timestamp markers into the audio-token stream. At the data level, we design an event-preserving audio annotation pipeline that segments raw audio at coherent event boundaries, applies branch-specific annotation to speech, music, and general audio, and merges the results into unified captions for pretraining. The intermediate branch-specific captions are further retained to support the construction of task-oriented SFT data. The model is pretrained on large-scale audio-language data, with time-aware objectives incorporated to support temporal grounding, and then undergoes multi-stage post-training to enhance instruction following and audio-grounded reasoning. We release 4B and 8B variants in both Instruct and Thinking configurations. MOSS-Audio achieves strong performance across general audio understanding, speech captioning, ASR, and timestamped ASR, positioning it as a promising understanding foundation for future voice agents.

Jiachen Zhang, Junyi Lao, Chenghao Liu et al.

Finite Element Analysis (FEA) serves as the cornerstone of modern engineering design. However, its workflow is inherently complex and relies heavily on domain expertise. Although recent efforts have integrated Large Language Models (LLMs) into FEA, existing approaches face limitations in handling multimodal inputs and executing complex tasks. To address these limitations, we propose VFEAgent, an end-to-end multi-agent system designed to automate FEA modeling and simulation directly from input images and problem descriptions. Our methodology integrates two core components: (1) a multimodal vision-language multi-agent pipeline that employs ReAct-driven reasoning to extract structured FEA specifications from heterogeneous inputs and (2) a verification-first code synthesis framework, incorporating robust self-debugging and fallback mechanisms to ensure executability and physical validity. We systematically evaluated the system across various engineering mechanics scenarios. The results demonstrate that VFEAgent achieves a high success rate in generating complete and physically valid simulations, outperforming LLM-based baseline methods in reliability and correctness. These findings validate the feasibility of automating the complete FEA workflow, highlighting the framework's potential to liberate engineers from tedious manual analysis.

Qiheng Mao, Zemin Liu, Chenghao Liu et al.

Recent studies have highlighted the limitations of message-passing based graph neural networks (GNNs), e.g., limited model expressiveness, over-smoothing, over-squashing, etc. To alleviate these issues, Graph Transformers (GTs) have been proposed which work in the paradigm that allows message passing to a larger coverage even across the whole graph. Hinging on the global range attention mechanism, GTs have shown a superpower for representation learning on homogeneous graphs. However, the investigation of GTs on heterogeneous information networks (HINs) is still under-exploited. In particular, on account of the existence of heterogeneity, HINs show distinct data characteristics and thus require different treatment. To bridge this gap, in this paper we investigate the representation learning on HINs with Graph Transformer, and propose a novel model named HINormer, which capitalizes on a larger-range aggregation mechanism for node representation learning. In particular, assisted by two major modules, i.e., a local structure encoder and a heterogeneous relation encoder, HINormer can capture both the structural and heterogeneous information of nodes on HINs for comprehensive node representations. We conduct extensive experiments on four HIN benchmark datasets, which demonstrate that our proposed model can outperform the state-of-the-art.

Fabian Latorre, Chenghao Liu, Doyen Sahoo et al.

Dynamic Time Warping (DTW) has become the pragmatic choice for measuring distance between time series. However, it suffers from unavoidable quadratic time complexity when the optimal alignment matrix needs to be computed exactly. This hinders its use in deep learning architectures, where layers involving DTW computations cause severe bottlenecks. To alleviate these issues, we introduce a new metric for time series data based on the Optimal Transport (OT) framework, called Optimal Transport Warping (OTW). OTW enjoys linear time/space complexity, is differentiable and can be parallelized. OTW enjoys a moderate sensitivity to time and shape distortions, making it ideal for time series. We show the efficacy and efficiency of OTW on 1-Nearest Neighbor Classification and Hierarchical Clustering, as well as in the case of using OTW instead of DTW in Deep Learning architectures.

Chenghao Liu, Xiaoyang Qu, Jianzong Wang et al.

Federated Learning (FL) has been widely concerned for it enables decentralized learning while ensuring data privacy. However, most existing methods unrealistically assume that the classes encountered by local clients are fixed over time. After learning new classes, this assumption will make the model's catastrophic forgetting of old classes significantly severe. Moreover, due to the limitation of communication cost, it is challenging to use large-scale models in FL, which will affect the prediction accuracy. To address these challenges, we propose a novel framework, Federated Enhanced Transformer (FedET), which simultaneously achieves high accuracy and low communication cost. Specifically, FedET uses Enhancer, a tiny module, to absorb and communicate new knowledge, and applies pre-trained Transformers combined with different Enhancers to ensure high precision on various tasks. To address local forgetting caused by new classes of new tasks and global forgetting brought by non-i.i.d (non-independent and identically distributed) class imbalance across different local clients, we proposed an Enhancer distillation method to modify the imbalance between old and new knowledge and repair the non-i.i.d. problem. Experimental results demonstrate that FedET's average accuracy on representative benchmark datasets is 14.1% higher than the state-of-the-art method, while FedET saves 90% of the communication cost compared to the previous method.

Jiachen Zhang, Junnan Nie, Junyi Lao et al.

Vision--language--action (VLA) policies are trained to imitate actions; their loss never asks them to estimate reward, progress, or future success. Their frozen representations nevertheless carry such information, and it can be read out and used to guide action choice without retraining the policy. From mixed successful and failed manipulation trajectories on LIBERO-Goal, we recover Monte-Carlo outcome targets using lightweight linear probes on frozen features. The targets are consistently predictable from OpenVLA, Pi0.5, DINOv2, and CLIP features, and substantially less so from baselines built on progress, time-to-go, task identity, or proprioception. To rule out task and temporal shortcuts, we evaluate the probes under same-task, same-timestep matched comparisons: Pi0.5 probes still reach roughly 92% pairwise ordering accuracy, while label-shuffled controls stay at chance. Used as a test-time selector over sampled Pi0.5 action prefixes, the same probe turns this offline finding into behavior: on push-plate, success rises from 26.7% under greedy decoding to 44.3%, with a second positive case on wine-rack. The gains are not universal and require additional inference compute, but the underlying finding is clean: frozen VLAs already encode information about success that their imitation objective never explicitly demands.

Mouxiang Chen, Chenghao Liu, Zemin Liu et al.

Unbiased learning to rank (ULTR) aims to train an unbiased ranking model from biased user click logs. Most of the current ULTR methods are based on the examination hypothesis (EH), which assumes that the click probability can be factorized into two scalar functions, one related to ranking features and the other related to bias factors. Unfortunately, the interactions among features, bias factors and clicks are complicated in practice, and usually cannot be factorized in this independent way. Fitting click data with EH could lead to model misspecification and bring the approximation error. In this paper, we propose a vector-based EH and formulate the click probability as a dot product of two vector functions. This solution is complete due to its universality in fitting arbitrary click functions. Based on it, we propose a novel model named Vectorization to adaptively learn the relevance embeddings and sort documents by projecting embeddings onto a base vector. Extensive experiments show that our method significantly outperforms the state-of-the-art ULTR methods on complex real clicks as well as simple simulated clicks.

Gerald Woo, Chenghao Liu, Akshat Kumar et al.

Deep learning for time series forecasting has traditionally operated within a one-model-per-dataset framework, limiting its potential to leverage the game-changing impact of large pre-trained models. The concept of universal forecasting, emerging from pre-training on a vast collection of time series datasets, envisions a single Large Time Series Model capable of addressing diverse downstream forecasting tasks. However, constructing such a model poses unique challenges specific to time series data: i) cross-frequency learning, ii) accommodating an arbitrary number of variates for multivariate time series, and iii) addressing the varying distributional properties inherent in large-scale data. To address these challenges, we present novel enhancements to the conventional time series Transformer architecture, resulting in our proposed Masked Encoder-based Universal Time Series Forecasting Transformer (Moirai). Trained on our newly introduced Large-scale Open Time Series Archive (LOTSA) featuring over 27B observations across nine domains, Moirai achieves competitive or superior performance as a zero-shot forecaster when compared to full-shot models. Code, data, and model weights can be found at https://github.com/SalesforceAIResearch/uni2ts.

Mouxiang Chen, Chenghao Liu, Zemin Liu et al.

Unbiased Learning to Rank (ULTR) aims to train unbiased ranking models from biased click logs, by explicitly modeling a generation process for user behavior and fitting click data based on examination hypothesis. Previous research found empirically that the true latent relevance is mostly recoverable through click fitting. However, we demonstrate that this is not always achievable, resulting in a significant reduction in ranking performance. This research investigates the conditions under which relevance can be recovered from click data in the first principle. We initially characterize a ranking model as identifiable if it can recover the true relevance up to a scaling transformation, a criterion sufficient for the pairwise ranking objective. Subsequently, we investigate an equivalent condition for identifiability, articulated as a graph connectivity test problem: the recovery of relevance is feasible if and only if the identifiability graph (IG), derived from the underlying structure of the dataset, is connected. The presence of a disconnected IG may lead to degenerate cases and suboptimal ranking performance. To tackle this challenge, we introduce two methods, namely node intervention and node merging, designed to modify the dataset and restore the connectivity of the IG. Empirical results derived from a simulated dataset and two real-world LTR benchmark datasets not only validate our proposed theory but also demonstrate the effectiveness of our methods in alleviating data bias when the relevance model is unidentifiable.

Xinru Zhang, Ni Ou, Chenghao Liu et al.

Automated brain tumor segmentation based on deep learning (DL) has achieved promising performance. However, it generally relies on annotated images for model training, which is not always feasible in clinical settings. Therefore, the development of unsupervised DL-based brain tumor segmentation approaches without expert annotations is desired. Motivated by the success of prompt learning (PL) in natural language processing, we propose an approach to unsupervised brain tumor segmentation by designing image-based prompts that allow indication of brain tumors, and this approach is dubbed as PL-based Brain Tumor Segmentation (PL-BTS). Specifically, instead of directly training a model for brain tumor segmentation with a large amount of annotated data, we seek to train a model that can answer the question: is a voxel in the input image associated with tumor-like hyper-/hypo-intensity? Such a model can be trained by artificially generating tumor-like hyper-/hypo-intensity on images without tumors with hand-crafted designs. Since the hand-crafted designs may be too simplistic to represent all kinds of real tumors, the trained model may overfit the simplistic hand-crafted task rather than actually answer the question of abnormality. To address this problem, we propose the use of a validation task, where we generate a different hand-crafted task to monitor overfitting. In addition, we propose PL-BTS+ that further improves PL-BTS by exploiting unannotated images with brain tumors. Compared with competing unsupervised methods, the proposed method has achieved marked improvements on both public and in-house datasets, and we have also demonstrated its possible extension to other brain lesion segmentation tasks.

Hongkai Li, Shifeng Xie, Lefei Shen et al.

Time series foundation models (TSFMs) are increasingly pretrained on large corpora, raising concerns that evaluation datasets may have been exposed during pretraining and thus yield overly optimistic performance estimates. Auditing such contamination is challenging in time series because signals are continuous and heterogeneous, and often lack corpus documentation. To the best of our knowledge, this is the first work to study pretraining contamination auditing for TSFMs. We formalize the problem of pretraining contamination auditing for TSFMs and propose TSFMAudit, a method based on probe adaptation dynamics. Our key intuition is that contamination manifests as unusually efficient adaptation: after a fine tuning probe, contaminated datasets tend to exhibit faster loss reduction with smaller backbone movement. We evaluate TSFMAudit on 6 TSFMs and 187 datasets using documented training source evidence as supervision, and compare against 10 competitive baselines adapted from the LLM literature.

Mouxiang Chen, Lefei Shen, Han Fu et al.

Recent years have witnessed the success of introducing deep learning models to time series forecasting. From a data generation perspective, we illustrate that existing models are susceptible to distribution shifts driven by temporal contexts, whether observed or unobserved. Such context-driven distribution shift (CDS) introduces biases in predictions within specific contexts and poses challenges for conventional training paradigms. In this paper, we introduce a universal calibration methodology for the detection and adaptation of CDS with a trained model. To this end, we propose a novel CDS detector, termed the "residual-based CDS detector" or "Reconditionor", which quantifies the model's vulnerability to CDS by evaluating the mutual information between prediction residuals and their corresponding contexts. A high Reconditionor score indicates a severe susceptibility, thereby necessitating model adaptation. In this circumstance, we put forth a straightforward yet potent adapter framework for model calibration, termed the "sample-level contextualized adapter" or "SOLID". This framework involves the curation of a contextually similar dataset to the provided test sample and the subsequent fine-tuning of the model's prediction layer with a limited number of steps. Our theoretical analysis demonstrates that this adaptation strategy can achieve an optimal bias-variance trade-off. Notably, our proposed Reconditionor and SOLID are model-agnostic and readily adaptable to a wide range of models. Extensive experiments show that SOLID consistently enhances the performance of current forecasting models on real-world datasets, especially on cases with substantial CDS detected by the proposed Reconditionor, thus validating the effectiveness of the calibration approach.

Chenghao Liu, Taha Aksu, Juncheng Liu et al.

We introduce Moirai 2.0, a decoder-only time-series foundation model trained on a new corpus of 36M series. The model adopts quantile forecasting and multi-token prediction, improving both probabilistic accuracy and inference efficiency. On the Gift-Eval benchmark, it ranks among the top pretrained models while achieving a strong trade-off between accuracy, speed, and model size. Compared to Moirai 1.0, Moirai 2.0 replaces masked-encoder training, multi-patch inputs, and mixture-distribution outputs with a simpler decoder-only architecture, single patch, and quantile loss. Ablation studies isolate these changes -- showing that the decoder-only backbone along with recursive multi-quantile decoding contribute most to the gains. Additional experiments show that Moirai 2.0 outperforms larger models from the same family and exhibits robust domain-level results. In terms of efficiency and model size, Moirai 2.0 is twice as fast and thirty times smaller than its prior best version, Moirai 1.0-Large, while also performing better. Model performance plateaus with increasing parameter count and declines at longer horizons, motivating future work on data scaling and long-horizon modeling. We release code and evaluation details to support further research.

Mouxiang Chen, Zemin Liu, Chenghao Liu et al.

Recent research has demonstrated the efficacy of pre-training graph neural networks (GNNs) to capture the transferable graph semantics and enhance the performance of various downstream tasks. However, the semantic knowledge learned from pretext tasks might be unrelated to the downstream task, leading to a semantic gap that limits the application of graph pre-training. To reduce this gap, traditional approaches propose hybrid pre-training to combine various pretext tasks together in a multi-task learning fashion and learn multi-grained knowledge, which, however, cannot distinguish tasks and results in some transferable task-specific knowledge distortion by each other. Moreover, most GNNs cannot distinguish nodes located in different parts of the graph, making them fail to learn position-specific knowledge and lead to suboptimal performance. In this work, inspired by the prompt-based tuning in natural language processing, we propose a unified framework for graph hybrid pre-training which injects the task identification and position identification into GNNs through a prompt mechanism, namely multi-task graph dual prompt (ULTRA-DP). Based on this framework, we propose a prompt-based transferability test to find the most relevant pretext task in order to reduce the semantic gap. To implement the hybrid pre-training tasks, beyond the classical edge prediction task (node-node level), we further propose a novel pre-training paradigm based on a group of $k$-nearest neighbors (node-group level). The combination of them across different scales is able to comprehensively express more structural semantics and derive richer multi-grained knowledge. Extensive experiments show that our proposed ULTRA-DP can significantly enhance the performance of hybrid pre-training methods and show the generalizability to other pre-training tasks and backbone architectures.

Taha Aksu, Gerald Woo, Juncheng Liu et al.

Time series foundation models excel in zero-shot forecasting, handling diverse tasks without explicit training. However, the advancement of these models has been hindered by the lack of comprehensive benchmarks. To address this gap, we introduce the General Time Series Forecasting Model Evaluation, GIFT-Eval, a pioneering benchmark aimed at promoting evaluation across diverse datasets. GIFT-Eval encompasses 23 datasets over 144,000 time series and 177 million data points, spanning seven domains, 10 frequencies, multivariate inputs, and prediction lengths ranging from short to long-term forecasts. To facilitate the effective pretraining and evaluation of foundation models, we also provide a non-leaking pretraining dataset containing approximately 230 billion data points. Additionally, we provide a comprehensive analysis of 17 baselines, which includes statistical models, deep learning models, and foundation models. We discuss each model in the context of various benchmark characteristics and offer a qualitative analysis that spans both deep learning and foundation models. We believe the insights from this analysis, along with access to this new standard zero-shot time series forecasting benchmark, will guide future developments in time series foundation models. Code, data, and the leaderboard can be found at https://github.com/SalesforceAIResearch/gift-eval .

Yingyuan Yang, Tian Lan, Yifei Gao et al.

Time-series anomaly detection (TSAD) requires identifying both immediate Point Anomalies and long-range Context Anomalies. However, existing foundation models face a fundamental trade-off: 1D temporal models provide fine-grained pointwise localization but lack a global contextual perspective, while 2D vision-based models capture global patterns but suffer from information bottlenecks due to a lack of temporal alignment and coarse-grained pointwise detection. To resolve this dilemma, we propose VETime, the first TSAD framework that unifies temporal and visual modalities through fine-grained visual-temporal alignment and dynamic fusion. VETime introduces a Reversible Image Conversion and a Patch-Level Temporal Alignment module to establish a shared visual-temporal timeline, preserving discriminative details while maintaining temporal sensitivity. Furthermore, we design an Anomaly Window Contrastive Learning mechanism and a Task-Adaptive Multi-Modal Fusion to adaptively integrate the complementary perceptual strengths of both modalities. Extensive experiments demonstrate that VETime significantly outperforms state-of-the-art models in zero-shot scenarios, achieving superior localization precision with lower computational overhead than current vision-based approaches. Code available at: https://github.com/yyyangcoder/VETime.

Emaad Khwaja, Chris Lettieri, Gerald Woo et al.

We show that time series foundation models scale: a single training recipe produces reliable forecast-quality improvements from 4M to 2.5B parameters. We release Toto 2.0, a family of five open-weights forecasting models trained under this recipe. The Toto 2.0 family sets a new state of the art on three forecasting benchmarks: BOOM, our observability benchmark; GIFT-Eval, the standard general-purpose benchmark; and the recent contamination-resistant TIME benchmark. This report describes our experimental results and details the design decisions behind Toto 2.0: its architecture and training recipe, training data, and the u-muP hyperparameter transfer pipeline. All five base checkpoints are released under Apache 2.0.

Giang Do, Khiem Le, Quang Pham et al.

By routing input tokens to only a few split experts, Sparse Mixture-of-Experts has enabled efficient training of large language models. Recent findings suggest that fixing the routers can achieve competitive performance by alleviating the collapsing problem, where all experts eventually learn similar representations. However, this strategy has two key limitations: (i) the policy derived from random routers might be sub-optimal, and (ii) it requires extensive resources during training and evaluation, leading to limited efficiency gains. This work introduces \HyperRout, which dynamically generates the router's parameters through a fixed hypernetwork and trainable embeddings to achieve a balance between training the routers and freezing them to learn an improved routing policy. Extensive experiments across a wide range of tasks demonstrate the superior performance and efficiency gains of \HyperRouter compared to existing routing methods. Our implementation is publicly available at {\url{https://github.com/giangdip2410/HyperRouter}}.

Chengzhi Liu, Zheng Tao, Zihong Luo et al.

Time series analysis and modelling constitute a crucial research area. Traditional artificial neural networks struggle with complex, non-stationary time series data due to high computational complexity, limited ability to capture temporal information, and difficulty in handling event-driven data. To address these challenges, we propose a Multi-modal Time Series Analysis Model Based on Spiking Neural Network (MTSA-SNN). The Pulse Encoder unifies the encoding of temporal images and sequential information in a common pulse-based representation. The Joint Learning Module employs a joint learning function and weight allocation mechanism to fuse information from multi-modal pulse signals complementary. Additionally, we incorporate wavelet transform operations to enhance the model's ability to analyze and evaluate temporal information. Experimental results demonstrate that our method achieved superior performance on three complex time-series tasks. This work provides an effective event-driven approach to overcome the challenges associated with analyzing intricate temporal information. Access to the source code is available at https://github.com/Chenngzz/MTSA-SNN}{https://github.com/Chenngzz/MTSA-SNN

Lefei Shen, Mouxiang Chen, Xu Liu et al.

Recent studies have indicated that vision models pre-trained on images can serve as time series foundation models (TSFMs) by reformulating time series forecasting (TSF) as image reconstruction. However, effective cross-modal transfer from vision to time series remains challenging due to three discrepancies: (1) the data-modality gap between structured, bounded image data and unbounded, heterogeneous time series; (2) the multivariate-forecasting gap between fixed RGB-three-channel vision models and time series with arbitrary numbers of variates; and (3) the probabilistic-forecasting gap between the deterministic outputs of vision models and the requirement for uncertainty-aware probabilistic predictions. To bridge these gaps, we propose VisonTS++, a TSFM based on continual pre-training of a vision model on large-scale time series. Our approach introduces three key innovations: (1) vision-model-based filtering to identify high-quality sequences to stabilize pre-training and mitigate modality gap; (2) colorized multivariate conversion, encoding multivariate series as multi-subfigure RGB images to enhance cross-variate modeling; (3) multi-quantile forecasting, using parallel reconstruction heads to generate quantile forecasts without parametric assumptions. Experiments show that VisionTS++ achieves state-of-the-art performance in both in-distribution and out-of-distribution forecasting, outperforming specialized TSFMs by 6%-44% in MSE reduction and ranking first in GIFT-Eval benchmark which comprises 23 datasets across 7 domains. Our work demonstrates that with appropriate adaptation, vision models can effectively generalize to TSF, thus advancing the pursuit of universal TSFMs. Code is available at https://github.com/HALF111/VisionTSpp.

Qingxiang Liu, Chenghao Liu, Sheng Sun et al.

Unsupervised anomaly detection of multivariate time series is a challenging task, given the requirements of deriving a compact detection criterion without accessing the anomaly points. The existing methods are mainly based on reconstruction error or association divergence, which are both confined to isolated subsequences with limited horizons, hardly promising unified series-level criterion. In this paper, we propose the Global Dictionary-enhanced Transformer (GDformer) with a renovated dictionary-based cross attention mechanism to cultivate the global representations shared by all normal points in the entire series. Accordingly, the cross-attention maps reflect the correlation weights between the point and global representations, which naturally leads to the representation-wise similarity-based detection criterion. To foster more compact detection boundary, prototypes are introduced to capture the distribution of normal point-global correlation weights. GDformer consistently achieves state-of-the-art unsupervised anomaly detection performance on five real-world benchmark datasets. Further experiments validate the global dictionary has great transferability among various datasets. The code is available at https://github.com/yuppielqx/GDformer.

Zhongzheng Qiao, Chenghao Liu, Yiming Zhang et al.

Time series foundation models (TSFMs) demonstrate impressive zero-shot performance for time series forecasting. However, an important yet underexplored challenge is how to effectively finetune TSFMs on specific downstream tasks. While naive finetuning can yield performance gains, we argue that it falls short of fully leveraging TSFMs' capabilities, often resulting in overfitting and suboptimal performance. Given the diverse temporal patterns across sampling scales and the inherent multi-scale forecasting capabilities of TSFMs, we adopt a causal perspective to analyze finetuning process, through which we highlight the critical importance of explicitly modeling multiple scales and reveal the shortcomings of naive approaches. Focusing on encoder-based TSFMs, we propose Multiscale finetuning (MSFT), a simple yet general framework that explicitly integrates multi-scale modeling into the finetuning process. Experimental results on three different backbones (Moirai, Moment and Units) demonstrate that TSFMs finetuned with MSFT not only outperform naive and typical parameter efficient finetuning methods but also surpass state-of-the-art deep learning methods. Codes are available at https://github.com/zqiao11/MSFT.

Junwei Deng, Chang Xu, Jiaqi W. Ma et al.

Time Series Foundation Models (TSFMs) are a powerful paradigm for time series analysis and are often enhanced by synthetic data augmentation to improve the training data quality. Existing augmentation methods, however, typically rely on heuristics and static paradigms. Motivated by dynamic data optimization, which shows that the contribution of samples varies across training stages, we propose OATS (Online Data Augmentation for Time Series Foundation Models), a principled strategy that generates synthetic data tailored to different training steps. OATS leverages valuable training samples as principled guiding signals and dynamically generates high-quality synthetic data conditioned on them. We further design a diffusion-based framework to produce realistic time series and introduce an explore-exploit mechanism to balance efficiency and effectiveness. Experiments on TSFMs demonstrate that OATS consistently outperforms regular training and yields substantial performance gains over static data augmentation baselines across six validation datasets and two TSFM architectures. The code is available at the link https://github.com/microsoft/TimeCraft.

Lefei Shen, Mouxiang Chen, Han Fu et al.

Transformer-based models have recently become dominant in Long-term Time Series Forecasting (LTSF), yet the variations in their architecture, such as encoder-only, encoder-decoder, and decoder-only designs, raise a crucial question: What Transformer architecture works best for LTSF tasks? However, existing models are often tightly coupled with various time-series-specific designs, making it difficult to isolate the impact of the architecture itself. To address this, we propose a novel taxonomy that disentangles these designs, enabling clearer and more unified comparisons of Transformer architectures. Our taxonomy considers key aspects such as attention mechanisms, forecasting aggregations, forecasting paradigms, and normalization layers. Through extensive experiments, we uncover several key insights: bi-directional attention with joint-attention is most effective; more complete forecasting aggregation improves performance; and the direct-mapping paradigm outperforms autoregressive approaches. Furthermore, our combined model, utilizing optimal architectural choices, consistently outperforms several existing models, reinforcing the validity of our conclusions. We hope these findings offer valuable guidance for future research on Transformer architectural designs in LTSF. Our code is available at https://github.com/HALF111/TSF_architecture.

Quang Pham, Chenghao Liu, Steven Hoi

Existing continual learning methods use Batch Normalization (BN) to facilitate training and improve generalization across tasks. However, the non-i.i.d and non-stationary nature of continual learning data, especially in the online setting, amplify the discrepancy between training and testing in BN and hinder the performance of older tasks. In this work, we study the cross-task normalization effect of BN in online continual learning where BN normalizes the testing data using moments biased towards the current task, resulting in higher catastrophic forgetting. This limitation motivates us to propose a simple yet effective method that we call Continual Normalization (CN) to facilitate training similar to BN while mitigating its negative effect. Extensive experiments on different continual learning algorithms and online scenarios show that CN is a direct replacement for BN and can provide substantial performance improvements. Our implementation is available at \url{https://github.com/phquang/Continual-Normalization}.

Quang Pham, Chenghao Liu, Doyen Sahoo et al.

The fast adaptation capability of deep neural networks in non-stationary environments is critical for online time series forecasting. Successful solutions require handling changes to new and recurring patterns. However, training deep neural forecaster on the fly is notoriously challenging because of their limited ability to adapt to non-stationary environments and the catastrophic forgetting of old knowledge. In this work, inspired by the Complementary Learning Systems (CLS) theory, we propose Fast and Slow learning Networks (FSNet), a holistic framework for online time-series forecasting to simultaneously deal with abrupt changing and repeating patterns. Particularly, FSNet improves the slowly-learned backbone by dynamically balancing fast adaptation to recent changes and retrieving similar old knowledge. FSNet achieves this mechanism via an interaction between two complementary components of an adapter to monitor each layer's contribution to the lost, and an associative memory to support remembering, updating, and recalling repeating events. Extensive experiments on real and synthetic datasets validate FSNet's efficacy and robustness to both new and recurring patterns. Our code is available at \url{https://github.com/salesforce/fsnet}.

Gerald Woo, Chenghao Liu, Doyen Sahoo et al.

Deep learning has been actively studied for time series forecasting, and the mainstream paradigm is based on the end-to-end training of neural network architectures, ranging from classical LSTM/RNNs to more recent TCNs and Transformers. Motivated by the recent success of representation learning in computer vision and natural language processing, we argue that a more promising paradigm for time series forecasting, is to first learn disentangled feature representations, followed by a simple regression fine-tuning step -- we justify such a paradigm from a causal perspective. Following this principle, we propose a new time series representation learning framework for time series forecasting named CoST, which applies contrastive learning methods to learn disentangled seasonal-trend representations. CoST comprises both time domain and frequency domain contrastive losses to learn discriminative trend and seasonal representations, respectively. Extensive experiments on real-world datasets show that CoST consistently outperforms the state-of-the-art methods by a considerable margin, achieving a 21.3% improvement in MSE on multivariate benchmarks. It is also robust to various choices of backbone encoders, as well as downstream regressors. Code is available at https://github.com/salesforce/CoST.

Gerald Woo, Chenghao Liu, Doyen Sahoo et al.

Transformers have been actively studied for time-series forecasting in recent years. While often showing promising results in various scenarios, traditional Transformers are not designed to fully exploit the characteristics of time-series data and thus suffer some fundamental limitations, e.g., they generally lack of decomposition capability and interpretability, and are neither effective nor efficient for long-term forecasting. In this paper, we propose ETSFormer, a novel time-series Transformer architecture, which exploits the principle of exponential smoothing in improving Transformers for time-series forecasting. In particular, inspired by the classical exponential smoothing methods in time-series forecasting, we propose the novel exponential smoothing attention (ESA) and frequency attention (FA) to replace the self-attention mechanism in vanilla Transformers, thus improving both accuracy and efficiency. Based on these, we redesign the Transformer architecture with modular decomposition blocks such that it can learn to decompose the time-series data into interpretable time-series components such as level, growth and seasonality. Extensive experiments on various time-series benchmarks validate the efficacy and advantages of the proposed method. Code is available at https://github.com/salesforce/ETSformer.

Aadyot Bhatnagar, Paul Kassianik, Chenghao Liu et al.

We introduce Merlion, an open-source machine learning library for time series. It features a unified interface for many commonly used models and datasets for anomaly detection and forecasting on both univariate and multivariate time series, along with standard pre/post-processing layers. It has several modules to improve ease-of-use, including visualization, anomaly score calibration to improve interpetability, AutoML for hyperparameter tuning and model selection, and model ensembling. Merlion also provides a unique evaluation framework that simulates the live deployment and re-training of a model in production. This library aims to provide engineers and researchers a one-stop solution to rapidly develop models for their specific time series needs and benchmark them across multiple time series datasets. In this technical report, we highlight Merlion's architecture and major functionalities, and we report benchmark numbers across different baseline models and ensembles.

Quang Pham, Doyen Sahoo, Chenghao Liu et al.

Continual learning aims to learn continuously from a stream of tasks and data in an online-learning fashion, being capable of exploiting what was learned previously to improve current and future tasks while still being able to perform well on the previous tasks. One common limitation of many existing continual learning methods is that they often train a model directly on all available training data without validation due to the nature of continual learning, thus suffering poor generalization at test time. In this work, we present a novel framework of continual learning named "Bilevel Continual Learning" (BCL) by unifying a {\it bilevel optimization} objective and a {\it dual memory management} strategy comprising both episodic memory and generalization memory to achieve effective knowledge transfer to future tasks and alleviate catastrophic forgetting on old tasks simultaneously. Our extensive experiments on continual learning benchmarks demonstrate the efficacy of the proposed BCL compared to many state-of-the-art methods. Our implementation is available at https://github.com/phquang/bilevel-continual-learning.

Yue Wu, Steven C. H. Hoi, Chenghao Liu et al.

SOL is an open-source library for scalable online learning algorithms, and is particularly suitable for learning with high-dimensional data. The library provides a family of regular and sparse online learning algorithms for large-scale binary and multi-class classification tasks with high efficiency, scalability, portability, and extensibility. SOL was implemented in C++, and provided with a collection of easy-to-use command-line tools, python wrappers and library calls for users and developers, as well as comprehensive documents for both beginners and advanced users. SOL is not only a practical machine learning toolbox, but also a comprehensive experimental platform for online learning research. Experiments demonstrate that SOL is highly efficient and scalable for large-scale machine learning with high-dimensional data.

Yiyuan Yang, Ming Jin, Haomin Wen et al.

The study of time series is crucial for understanding trends and anomalies over time, enabling predictive insights across various sectors. Spatio-temporal data, on the other hand, is vital for analyzing phenomena in both space and time, providing a dynamic perspective on complex system interactions. Recently, diffusion models have seen widespread application in time series and spatio-temporal data mining. Not only do they enhance the generative and inferential capabilities for sequential and temporal data, but they also extend to other downstream tasks. In this survey, we comprehensively and thoroughly review the use of diffusion models in time series and spatio-temporal data, categorizing them by model category, task type, data modality, and practical application domain. In detail, we categorize diffusion models into unconditioned and conditioned types and discuss time series and spatio-temporal data separately. Unconditioned models, which operate unsupervised, are subdivided into probability-based and score-based models, serving predictive and generative tasks such as forecasting, anomaly detection, classification, and imputation. Conditioned models, on the other hand, utilize extra information to enhance performance and are similarly divided for both predictive and generative tasks. Our survey extensively covers their application in various fields, including healthcare, recommendation, climate, energy, audio, and transportation, providing a foundational understanding of how these models analyze and generate data. Through this structured overview, we aim to provide researchers and practitioners with a comprehensive understanding of diffusion models for time series and spatio-temporal data analysis, aiming to direct future innovations and applications by addressing traditional challenges and exploring innovative solutions within the diffusion model framework.

Xu Liu, Juncheng Liu, Gerald Woo et al.

Time series foundation models have demonstrated impressive performance as zero-shot forecasters. However, achieving effectively unified training on time series remains an open challenge. Existing approaches introduce some level of model specialization to account for the highly heterogeneous nature of time series data. For instance, Moirai pursues unified training by employing multiple input/output projection layers, each tailored to handle time series at a specific frequency. Similarly, TimesFM maintains a frequency embedding dictionary for this purpose. We identify two major drawbacks to this human-imposed frequency-level model specialization: (1) Frequency is not a reliable indicator of the underlying patterns in time series. For example, time series with different frequencies can display similar patterns, while those with the same frequency may exhibit varied patterns. (2) Non-stationarity is an inherent property of real-world time series, leading to varied distributions even within a short context window of a single time series. Frequency-level specialization is too coarse-grained to capture this level of diversity. To address these limitations, this paper introduces Moirai-MoE, using a single input/output projection layer while delegating the modeling of diverse time series patterns to the sparse mixture of experts (MoE) within Transformers. With these designs, Moirai-MoE reduces reliance on human-defined heuristics and enables automatic token-level specialization. Extensive experiments on 39 datasets demonstrate the superiority of Moirai-MoE over existing foundation models in both in-distribution and zero-shot scenarios. Furthermore, this study conducts comprehensive model analyses to explore the inner workings of time series MoE foundation models and provides valuable insights for future research.

Quang Pham, Giang Do, Huy Nguyen et al.

Sparse mixture of experts (SMoE) offers an appealing solution to scale up the model complexity beyond the mean of increasing the network's depth or width. However, effective training of SMoE has proven to be challenging due to the representation collapse issue, which causes parameter redundancy and limited representation potentials. In this work, we propose a competition mechanism to address this fundamental challenge of representation collapse. By routing inputs only to experts with the highest neural response, we show that, under mild assumptions, competition enjoys the same convergence rate as the optimal estimator. We further propose CompeteSMoE, an effective and efficient algorithm to train large language models by deploying a simple router that predicts the competition outcomes. Consequently, CompeteSMoE enjoys strong performance gains from the competition routing policy while having low computation overheads. Our extensive empirical evaluations on two transformer architectures and a wide range of tasks demonstrate the efficacy, robustness, and scalability of CompeteSMoE compared to state-of-the-art SMoE strategies.

Qingxiang Liu, Xu Liu, Chenghao Liu et al.

Unlike natural language processing and computer vision, the development of Foundation Models (FMs) for time series forecasting is blocked due to data scarcity. While recent efforts are focused on building such FMs by unlocking the potential of language models (LMs) for time series analysis, dedicated parameters for various downstream forecasting tasks need training, which hinders the common knowledge sharing across domains. Moreover, data owners may hesitate to share the access to local data due to privacy concerns and copyright protection, which makes it impossible to simply construct a FM on cross-domain training instances. To address these issues, we propose Time-FFM, a Federated Foundation Model for Time series forecasting by leveraging pretrained LMs. Specifically, we begin by transforming time series into the modality of text tokens. To bootstrap LMs for time series reasoning, we propose a prompt adaption module to determine domain-customized prompts dynamically instead of artificially. Given the data heterogeneity across domains, we design a personalized federated training strategy by learning global encoders and local prediction heads. Our comprehensive experiments indicate that Time-FFM outperforms state-of-the-arts and promises effective few-shot and zero-shot forecaster.

Haoxin Liu, Chenghao Liu, B. Aditya Prakash

Large language models (LLMs), with demonstrated reasoning abilities across multiple domains, are largely underexplored for time-series reasoning (TsR), which is ubiquitous in the real world. In this work, we propose TimerBed, the first comprehensive testbed for evaluating LLMs' TsR performance. Specifically, TimerBed includes stratified reasoning patterns with real-world tasks, comprehensive combinations of LLMs and reasoning strategies, and various supervised models as comparison anchors. We perform extensive experiments with TimerBed, test multiple current beliefs, and verify the initial failures of LLMs in TsR, evidenced by the ineffectiveness of zero shot (ZST) and performance degradation of few shot in-context learning (ICL). Further, we identify one possible root cause: the numerical modeling of data. To address this, we propose a prompt-based solution VL-Time, using visualization-modeled data and language-guided reasoning. Experimental results demonstrate that Vl-Time enables multimodal LLMs to be non-trivial ZST and powerful ICL reasoners for time series, achieving about 140% average performance improvement and 99% average token costs reduction.

Yaxuan Kong, Yiyuan Yang, Shiyu Wang et al.

Understanding time series data is crucial for multiple real-world applications. While large language models (LLMs) show promise in time series tasks, current approaches often rely on numerical data alone, overlooking the multimodal nature of time-dependent information, such as textual descriptions, visual data, and audio signals. Moreover, these methods underutilize LLMs' reasoning capabilities, limiting the analysis to surface-level interpretations instead of deeper temporal and multimodal reasoning. In this position paper, we argue that multimodal LLMs (MLLMs) can enable more powerful and flexible reasoning for time series analysis, enhancing decision-making and real-world applications. We call on researchers and practitioners to leverage this potential by developing strategies that prioritize trust, interpretability, and robust reasoning in MLLMs. Lastly, we highlight key research directions, including novel reasoning paradigms, architectural innovations, and domain-specific applications, to advance time series reasoning with MLLMs.

Qiheng Mao, Zemin Liu, Chenghao Liu et al.

The integration of Large Language Models (LLMs) with Graph Representation Learning (GRL) marks a significant evolution in analyzing complex data structures. This collaboration harnesses the sophisticated linguistic capabilities of LLMs to improve the contextual understanding and adaptability of graph models, thereby broadening the scope and potential of GRL. Despite a growing body of research dedicated to integrating LLMs into the graph domain, a comprehensive review that deeply analyzes the core components and operations within these models is notably lacking. Our survey fills this gap by proposing a novel taxonomy that breaks down these models into primary components and operation techniques from a novel technical perspective. We further dissect recent literature into two primary components including knowledge extractors and organizers, and two operation techniques including integration and training stratigies, shedding light on effective model design and training strategies. Additionally, we identify and explore potential future research avenues in this nascent yet underexplored field, proposing paths for continued progress.

Zhisheng Lin, Han Fu, Chenghao Liu et al.

Parameter-efficient fine-tuning (PEFT) has emerged as an effective method for adapting pre-trained language models to various tasks efficiently. Recently, there has been a growing interest in transferring knowledge from one or multiple tasks to the downstream target task to achieve performance improvements. However, current approaches typically either train adapters on individual tasks or distill shared knowledge from source tasks, failing to fully exploit task-specific knowledge and the correlation between source and target tasks. To overcome these limitations, we propose PEMT, a novel parameter-efficient fine-tuning framework based on multi-task transfer learning. PEMT extends the mixture-of-experts (MoE) framework to capture the transferable knowledge as a weighted combination of adapters trained on source tasks. These weights are determined by a gated unit, measuring the correlation between the target and each source task using task description prompt vectors. To fully exploit the task-specific knowledge, we also propose the Task Sparsity Loss to improve the sparsity of the gated unit. We conduct experiments on a broad range of tasks over 17 datasets. The experimental results demonstrate our PEMT yields stable improvements over full fine-tuning, and state-of-the-art PEFT and knowledge transferring methods on various tasks. The results highlight the effectiveness of our method which is capable of sufficiently exploiting the knowledge and correlation features across multiple tasks.