Deep learning · foundation models

Times series foundation models (TSFMs) have emerged as a promising paradigm for time series analysis and forecasting, showing remarkable generalization performance across different domains. While efforts have been made on hallucinations of foundation models, the hallucinations of TSFMs have been underexplored. In this paper, we formally define TSFM hallucinations in the zero-shot forecasting setting by examining whether a generated forecast exhibits different dynamics from those of the context. Our study reveals that TSFM hallucinations primarily stem from the loss of context information in hidden states during forward propagation. As such, we propose methodologies to identify signal subspaces in TSFMs and magnify their information through intervention. Extensive experiments demonstrate that our proposed intervention approach effectively mitigates hallucinations and improves forecasting performance. Furthermore, the signal strength measure we compute from signal subspaces has strong predictive power of hallucinations and forecasting performance of the model. Our work contributes to deeper understanding of TSFM trustworthiness that could foster future research in this direction.

Since the seminal work of TabPFN, research on tabular foundation models (TFMs) based on in-context learning (ICL) has challenged long-standing paradigms in machine learning. Without seeing any real-world data, models pretrained on purely synthetic datasets generalize remarkably well across diverse datasets, often using only a moderate number of in-context examples. This shifts the focus in tabular machine learning from model architecture design to the design of synthetic datasets, or, more precisely, to the prior distributions that generate them. Yet the guiding principles for prior design remain poorly understood. This work marks the first attempt to address the gap. We systematically investigate and identify key properties of synthetic priors that allow pretrained TFMs to generalize well. Based on these insights, we introduce Mitra, a TFM trained on a curated mixture of synthetic priors selected for their diversity, distinctiveness, and performance on real-world tabular data. Mitra consistently outperforms state-of-the-art TFMs, such as TabPFNv2 and TabICL, across both classification and regression benchmarks, with better sample efficiency.

The advent of single-cell Assay for Transposase-Accessible Chromatin using sequencing (scATAC-seq) offers an innovative perspective for deciphering regulatory mechanisms by assembling a vast repository of single-cell chromatin accessibility data. While foundation models have achieved significant success in single-cell transcriptomics, there is currently no foundation model for scATAC-seq that supports zero-shot high-quality cell identification and comprehensive multi-omics analysis simultaneously. Key challenges lie in the high dimensionality and sparsity of scATAC-seq data, as well as the lack of a standardized schema for representing open chromatin regions (OCRs). Here, we present ChromFound, a foundation model tailored for scATAC-seq. ChromFound utilizes a hybrid architecture and genome-aware tokenization to effectively capture genome-wide long contexts and regulatory signals from dynamic chromatin landscapes. Pretrained on 1.97 million cells from 30 tissues and 6 disease conditions, ChromFound demonstrates broad applicability across 6 diverse tasks. Notably, it achieves robust zero-shot performance in generating universal cell representations and exhibits excellent transferability in cell type annotation and cross-omics prediction. By uncovering enhancer-gene links undetected by existing computational methods, ChromFound offers a promising framework for understanding disease risk variants in the noncoding genome. The implementation of ChromFound is available via https://github.com/JohnsonKlose/ChromFound.

Scaling laws motivate the development of Time Series Foundation Models (TSFMs) that pre-train vast parameters and achieve remarkable zero-shot forecasting performance. Surprisingly, even after fine-tuning, TSFMs cannot consistently outperform smaller, specialized models trained on full-shot downstream data. A key question is how to realize effective adaptation of TSFMs for a target forecasting task. Through empirical studies on various TSFMs, the pre-trained models often exhibit inherent sparsity and redundancy in computation, suggesting that TSFMs have learned to activate task-relevant network substructures to accommodate diverse forecasting tasks. To preserve this valuable prior knowledge, we propose a structured pruning method to regularize the subsequent fine-tuning process by focusing it on a more relevant and compact parameter space. Extensive experiments on seven TSFMs and six benchmarks demonstrate that fine-tuning a smaller, pruned TSFM significantly improves forecasting performance compared to fine-tuning original models. This ``prune-then-finetune'' paradigm often enables TSFMs to achieve state-of-the-art performance and surpass strong specialized baselines. Source code is made publicly available at \url{https://github.com/SJTU-DMTai/Prune-then-Finetune}.

Spatio-Temporal Foundation Models (STFMs) promise zero/few-shot generalization across various datasets, yet joint spatio-temporal pretraining is computationally prohibitive and struggles with domain-specific spatial correlations. To this end, we introduce FactoST, a factorized STFM that decouples universal temporal pretraining from spatio-temporal adaptation. The first stage pretrains a space-agnostic backbone with multi-frequency reconstruction and domain-aware prompting, capturing cross-domain temporal regularities at low computational cost. The second stage freezes or further fine-tunes the backbone and attaches an adapter that fuses spatial metadata, sparsifies interactions, and aligns domains with continual memory replay. Extensive forecasting experiments reveal that, in few-shot setting, FactoST reduces MAE by up to 46.4% versus UniST, uses 46.2% fewer parameters, and achieves 68% faster inference than OpenCity, while remaining competitive with expert models. We believe this factorized view offers a practical and scalable path toward truly universal STFMs. The code will be released upon notification.

The advent of foundation models in AI has significantly advanced general-purpose learning, enabling remarkable capabilities in zero-shot inference and in-context learning. However, training such models on physics data, including solutions to partial differential equations (PDEs), poses a unique challenge due to varying dimensionalities across different systems. Traditional approaches either fix a maximum dimension or employ separate encoders for different dimensionalities, resulting in inefficiencies. To address this, we propose a dimension-agnostic neural network architecture, the Axial Neural Network (XNN), inspired by parameter-sharing structures such as Deep Sets and Graph Neural Networks. XNN generalizes across varying tensor dimensions while maintaining computational efficiency. We convert existing PDE foundation models into axial neural networks and evaluate their performance across three training scenarios: training from scratch, pretraining on multiple PDEs, and fine-tuning on a single PDE. Our experiments show that XNNs perform competitively with original models and exhibit superior generalization to unseen dimensions, highlighting the importance of multidimensional pretraining for foundation models.

Understanding and decoding human brain activity from electroencephalography (EEG) signals is a fundamental problem in neuroscience and artificial intelligence, with applications ranging from cognition and emotion recognition to clinical diagnosis and brain–computer interfaces. While recent EEG foundation models have made progress in generalized brain decoding by leveraging unified architectures and large-scale pretraining, they inherit a scale-agnostic dense modeling paradigm from NLP and vision. This design overlooks an intrinsic property of neural activity—cross-scale spatiotemporal structure. Different EEG task patterns span a broad range of temporal and spatial scales, from brief neural activations to slow-varying rhythms, and from localized cortical activations to large-scale distributed interactions. Ignoring this diversity may lead to suboptimal representations and weakened generalization ability. To address these limitations, we propose CSBrain, a Cross-scale Spatiotemporal Brain foundation model for generalized EEG decoding. CSBrain introduces two key components: (i) Cross-scale Spatiotemporal Tokenization (CST), which aggregates multi-scale features within localized temporal windows and anatomical brain regions into compact scale-aware token representations; and (ii) Structured Sparse Attention (SSA), which models cross-window and cross-region dependencies for diverse decoding tasks, further enriching scale diversities while eliminating the spurious dependencies. CST and SSA are alternately stacked to progressively integrate cross-scale spatiotemporal dependencies. Extensive experiments across 11 representative EEG tasks and 16 datasets demonstrate that CSBrain consistently outperforms both task-specific models and strong foundation baselines. These results establish cross-scale modeling as a key inductive bias for generalized EEG decoding and highlight CSBrain as a robust backbone for future brain–AI research.

Despite the remarkable successes of general-purpose neural networks, such as MLPs and Transformers, we find that they exhibit notable shortcomings in modeling and reasoning about periodic phenomena, achieving only marginal performance within the training domain and failing to generalize effectively to out-of-domain (OOD) scenarios. Periodicity is ubiquitous throughout nature and science. Therefore, neural networks should be equipped with the essential ability to model and handle periodicity. In this work, we propose FAN, a novel neural network that effectively addresses periodicity modeling challenges while offering broad applicability similar to MLP with fewer parameters and FLOPs. Periodicity is naturally integrated into FAN's structure and computational processes by introducing the Fourier Principle. Unlike existing Fourier-based networks, which possess particular periodicity modeling abilities but face challenges in scaling to deeper networks and are typically designed for specific tasks, our approach overcomes this challenge to enable scaling to large-scale models and maintains the capability to be applied to more types of tasks. Through extensive experiments, we demonstrate the superiority of FAN in periodicity modeling tasks and the effectiveness and generalizability of FAN across a range of real-world tasks. Moreover, we reveal that compared to existing Fourier-based networks, FAN accommodates both periodicity modeling and general-purpose modeling well.

We propose BlockScan, a customized Transformer for anomaly detection in blockchain transactions. Unlike existing methods that rely on rule-based systems or directly apply off-the-shelf large language models (LLMs), BlockScan introduces a series of customized designs to effectively model the unique data structure of blockchain transactions. First, a blockchain transaction is multi-modal, containing blockchain-specific tokens, texts, and numbers. We design a novel modularized tokenizer to handle these multi-modal inputs, balancing the information across different modalities. Second, we design a customized masked language modeling mechanism for pretraining the Transformer architecture, incorporating RoPE embedding and FlashAttention for handling longer sequences. Finally, we design a novel anomaly detection method based on the model outputs. We further provide theoretical analysis for the detection method of our system. Extensive evaluations on Ethereum and Solana transactions demonstrate BlockScan's exceptional capability in anomaly detection while maintaining a low false positive rate. Remarkably, BlockScan is the only method that successfully detects anomalous transactions on Solana with high accuracy, whereas all other approaches achieved very low or zero detection recall scores. This work sets a new benchmark for applying Transformer-based approaches in blockchain data analysis.

This work presents a systematic investigation into the latent knowledge encoded within Network Foundation Models (NFMs). Different from existing efforts, we focus on hidden representations analysis rather than pure downstream task performance and analyze NFMs through a three-part evaluation: Embedding Geometry Analysis to assess representation space utilization, Metric Alignment Assessment to measure correspondence with domain-expert features, and Causal Sensitivity Testing to evaluate robustness to protocol perturbations. Using five diverse network datasets spanning controlled and real-world environments, we evaluate four state-of-the-art NFMs, revealing that they all exhibit significant anisotropy, inconsistent feature sensitivity patterns, an inability to separate the high-level context, payload dependency, and other properties. Our work identifies numerous limitations across all models and demonstrates that addressing them can significantly improve model performance (up to 0.35 increase in $F_1$ scores without architectural changes).

Foundation models trained on large-scale de-identified electronic health records (EHRs) hold promise for clinical applications. However, their capacity to memorize patient information raises important privacy concerns. In this work, we introduce a suite of black-box evaluation tests to assess privacy-related memorization risks in foundation models trained on structured EHR data. Our framework includes methods for probing memorization at both the embedding and generative levels, and aims to distinguish between model generalization and harmful memorization in clinically relevant settings. We contextualize memorization in terms of its potential to compromise patient privacy, particularly for vulnerable subgroups. We validate our approach on a publicly available EHR foundation model and release an open-source toolkit to facilitate reproducible and collaborative privacy assessments in healthcare AI.

Recent foundational models for tabular data, such as TabPFN, excel at adapting to new tasks via in-context learning but remain constrained to a fixed, pre-defined number of target dimensions—often necessitating costly ensembling strategies. We trace this constraint to a deeper architectural shortcoming: these models lack target-equivariance, so that permuting target-dimension orderings alters their predictions. This deficiency gives rise to an irreducible “equivariance gap,” an error term that introduces instability in predictions. We eliminate this gap by designing a fully target-equivariant architecture—ensuring permutation invariance via equivariant encoders, decoders, and a bi-attention mechanism. Empirical evaluation on standard classification benchmarks shows that, on datasets with more classes than those seen during pre-training, our model matches or surpasses existing methods while incurring lower computational overhead.

Building a universal trajectory foundation model is a promising solution to address the limitations of existing trajectory modeling approaches, such as task specificity, regional dependency, and data sensitivity. Despite its potential, data preparation, pre-training strategy development, and architectural design present significant challenges in constructing this model. Therefore, we introduce **UniTraj**, a Universal Trajectory foundation model that aims to address these limitations through three key innovations. First, we construct **WorldTrace**, an unprecedented dataset of 2.45 million trajectories with billions of GPS points spanning 70 countries, providing the diverse geographic coverage essential for region-independent modeling. Second, we develop novel pre-training strategies--Adaptive Trajectory Resampling and Self-supervised Trajectory Masking--that enable robust learning from heterogeneous trajectory data with varying sampling rates and quality. Finally, we tailor a flexible model architecture to accommodate a variety of trajectory tasks, effectively capturing complex movement patterns to support broad applicability. Extensive experiments across multiple tasks and real-world datasets demonstrate that UniTraj consistently outperforms existing methods, exhibiting superior scalability, adaptability, and generalization, with WorldTrace serving as an ideal yet non-exclusive training resource. The implementation codes and full dataset are available at https://github.com/Yasoz/UniTraj.

Electroencephalography (EEG) has wide-ranging applications, from clinical diagnosis to brain-computer interfaces (BCIs). With the increasing volume and variety of EEG data, there has been growing interest in establishing foundation models (FMs) to scale up and generalize neural decoding. Despite showing early potential, applying FMs to EEG remains challenging due to substantial inter-subject, inter-task, and inter-condition variability, as well as diverse electrode configurations across recording setups. To tackle these open challenges, we propose **NeurIPT**, a foundation model tailored for diverse EEG-based **Neur**al **I**nterfaces with a **P**re-trained **T**ransformer by capturing both homogeneous and heterogeneous spatio-temporal characteristics inherent in EEG signals. Temporally, we introduce Amplitude-Aware Masked Pretraining (AAMP), masking based on signal amplitude rather than random intervals, to learn robust representations across varying signal intensities beyond local interpolation. Moreover, this temporal representation is enhanced by a progressive Mixture-of-Experts (MoE) architecture, where specialized expert subnetworks are progressively introduced at deeper layers, adapting effectively to the diverse temporal characteristics of EEG signals. Spatially, NeurIPT leverages the 3D physical coordinates of electrodes, enabling effective transfer across varying EEG settings, and develops Intra-Inter Lobe Pooling (IILP) during fine-tuning to efficiently exploit regional brain features. Empirical evaluations across nine downstream BCI datasets, via fine-tuning and training from scratch, demonstrated NeurIPT consistently achieved state-of-the-art performance, highlighting its broad applicability and robust generalization. Our work pushes forward the state of FMs in EEG and offers insights into scalable and generalizable neural information processing systems.

Large language models (LLMs) have revolutionized natural language processing and are increasingly applied to other sequential data types, including genetic sequences. However, adapting LLMs to genetics presents significant challenges. Capturing complex genomic interactions requires modeling long-range global dependencies within DNA sequences, where interactions often span over 10,000 base pairs, even within a single gene. This poses substantial computational demands under conventional model architectures and training paradigms. Additionally, traditional LLM training approaches are suboptimal for DNA sequences: autoregressive training, while efficient for training, only supports unidirectional sequence understanding. However, DNA is inherently bidirectional. For instance, bidirectional promoters regulate gene expression in both directions and govern approximately 11% of human gene expression. Masked language models (MLMs) enable bidirectional understanding. However, they are inefficient since only masked tokens contribute to loss calculations at each training step. To address these limitations, we introduce JanusDNA, the first bidirectional DNA foundation model built upon a novel pretraining paradigm, integrating the optimization efficiency of autoregressive modeling with the bidirectional comprehension capability of masked modeling. JanusDNA's architecture leverages a Mamba-Attention Mixture-of-Experts (MoE) design, combining the global, high-resolution context awareness of attention mechanisms with the efficient sequential representation learning capabilities of Mamba. The MoE layers further enhance the model's capacity through sparse parameter scaling, while maintaining manageable computational costs. Notably, JanusDNA can process up to 1 million base pairs at single-nucleotide resolution on a single 80GB GPU using its hybrid architecture. Extensive experiments and ablation studies demonstrate that JanusDNA achieves new state-of-the-art performance on three genomic representation benchmarks. Remarkably, JanusDNA surpasses models with 250x more activated parameters, underscoring its efficiency and effectiveness. Code available at https://anonymous.4open.science/r/JanusDNA/.