Technical Papers Archive

ReLATE: Accelerating Tensor Decomposition via Safe and Efficient Learning of Sparse Encodings


Authors: Ahmed E. Helal and Fabio Checconi (Intel Corporation), Jan Laukemann (FAU Erlangen-Nürnberg), Yongseok Soh (University of Oregon), Jesmin Jahan Tithi and Fabrizio Petrini (Intel Corporation), and Jee W. Choi (University of Oregon)

Abstract: Tensor decomposition (TD) is essential for analyzing high-dimensional sparse data, yet its irregular computations and memory-access patterns pose major performance challenges on modern parallel processors. Prior works rely on expert-designed sparse tensor formats that fail to adapt to irregular tensor shapes and data distributions. We present the reinforcement-learned adaptive tensor encoding (ReLATE) framework, a learning-augmented method that discovers safe and efficient sparse encodings, without labeled examples, via a hybrid model-free and model-based algorithm that learns from both real and imagined actions. Moreover, ReLATE introduces elastic training, rule-driven action masking, and dynamics-informed action filtering to ensure correct encoding with bounded execution time, even during early learning. After offline training, with geometric-mean overhead of only 5.82% relative to TD workflow time, ReLATE deploys the best encoding with zero inference overhead. Across diverse real-world sparse tensors, ReLATE consistently outperforms the best expert-designed format by up to 2×, with a geometric-mean speedup of 1.38×–1.41×.



Back to Technical Papers Archive Listing