Tiny OTIS encoder achieves SOTA time series feature learning

Researchers have developed OTIS, a novel open time series encoder designed for efficient deployment on resource-constrained systems. Despite its small size (7.1M parameters), OTIS achieves state-of-the-art performance across 162 tasks, outperforming much larger encoders by a significant margin in terms of memory, energy, and latency. The system incorporates a domain-aware tokenizer, a dual masking strategy, and a structure-aware objective to effectively learn time series features without the need for massive scale. The code and pre-trained weights for OTIS have been released to promote broader accessibility. AI

IMPACT Enables high-quality time series analysis on resource-constrained devices, democratizing advanced AI capabilities.

RANK_REASON The cluster contains an academic paper detailing a new model/methodology. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• OTIS
• Özgün Turgut

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →