Professor Rafael Ballester-Ripoll recently presented his latest research at the Thirty-Ninth AAAI Conference on Artificial Intelligence (AAAI-25) during the CoLoRAI Workshop on Compact Low-Rank AI Representations, held on March 4, 2025, in Philadelphia, USA. His talk, titled “Entropy Coding Compression of Tree Tensor Networks”, introduced a novel approach to tensor compression, extending beyond traditional Tucker decomposition methods to improve data reduction for storage and transmission.

His research explored the use of entropy coding and successive core orthogonalization applied to SVD-learned coefficients, leading to significant efficiency gains in tensor compression. Unlike conventional low-rank tensor decompositions, which primarily focus on compact representation for machine learning tasks, this approach prioritizes optimized storage and data transmission. By generalizing beyond Tucker-based methods to more general acyclic tensor networks, the proposed technique broadens its applicability to diverse datasets, from neural network activations to medical imaging and physical simulations.

The CoLoRAI Workshop brought together researchers working on structured low-rank representations in AI, fostering discussions on their applications in deep learning, probabilistic inference, and quantum computing. Professor Ballester-Ripoll’s contribution helped advance the dialogue on bridging different tensor-based methodologies, further reinforcing IE Research Datalab’s presence in cutting-edge AI research.