2 papers
cs.LG2026
Dynamic Tokenization via Reinforcement Patching: End-to-end Training and Zero-shot Transfer
Yulun Wu, Sravan Kumar Ankireddy, Samuel Sharpe +4
Efficiently aggregating spatial or temporal horizons to acquire compact representations has become a unifying principle in modern deep learning models, yet learning data-adaptive r…
cs.AI2026
TimeSqueeze: Dynamic Patching for Efficient Time Series Forecasting
Sravan Kumar Ankireddy, Nikita Seleznev, Nam H. Nguyen +4
Transformer-based time series foundation models face a fundamental trade-off in choice of tokenization: point-wise embeddings preserve temporal fidelity but scale poorly with seque…