Tabular foundation models excel on tiny- to medium-sized IID data but are outperformed by traditional tree-based and deep learning models on non-IID, large, and high-dimensional datasets, based on evaluations across 11 models and 142 datasets in the new BeyondArena benchmark.
Mantis: Lightweight calibrated foundation model for user-friendly time series classification
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
abstract
While foundation models have revolutionized various domains, their application to time series classification remains rather under-explored, with existing literature predominantly focused on forecasting. To bridge this gap, we introduce \textbf{Mantis}, a transformer-based foundation model pre-trained exclusively on synthetic data via self-supervised contrastive learning. We demonstrate that effective tokenization is critical to unlocking the full potential of transformers, proposing a novel token generator unit. Furthermore, we introduce an enhanced test-time methodology that bridges the performance gap between Mantis and strong specialized approaches by leveraging intermediate-layer representations, self-ensembling, and cross-model embedding fusion. Extensive experiments demonstrate that Mantis establishes a new state-of-the-art, outperforming existing foundation models across four diverse dataset collections covering various application domains.
citation-role summary
baseline 1
citation-polarity summary
roles
baseline 1polarities
baseline 1representative citing papers
A causal DAG prior synthesizes labeled multivariate TSC datasets with temporal cross-modal structure, yielding statistically significant gains when finetuning TabPFN v2.5 on 75 UCR/UEA datasets over unmodified and tabular-only baselines.
RocketPFN matches the accuracy of the strongest time series classifier HC2 on 92 UCR datasets using a training-free pipeline of Rocket features and TabPFN.
FMplex is a serving system that virtualizes FM backbones for sharing across tasks, claiming up to 80% lower latency and 6x more tasks hosted versus prior approaches.
MoRA is a new retrieval-augmented module for IMU-based human activity recognition that uses uncertainty-adaptive fusion of retrieved motion patterns to improve model performance.
COMODO is a cross-modal self-supervised distillation framework that uses a frozen video encoder and dynamic instance queue to align video and IMU embeddings, improving IMU-based egocentric HAR to match supervised performance.
LeNEPA proposes a no-augmentation next-latent prediction recipe that maintains frozen-probe performance across ECG and synthetic diagnostic time-series datasets under fixed-recipe conditions where a tuned JEPA baseline degrades.
citing papers explorer
No citing papers match the current filters.