Christodoulos Kechris
@chriskechris.bsky.social
PhD researcher at Embedded Systems Lab - EPFL
Project page: eslweb.epfl.ch/time-series-...
Paper: arxiv.org/pdf/2505.13100
Code (PyTorch/Captum/TensorFlow): github.com/esl-epfl/cro...
Paper: arxiv.org/pdf/2505.13100
Code (PyTorch/Captum/TensorFlow): github.com/esl-epfl/cro...
May 19, 2026 at 9:27 AM
Project page: eslweb.epfl.ch/time-series-...
Paper: arxiv.org/pdf/2505.13100
Code (PyTorch/Captum/TensorFlow): github.com/esl-epfl/cro...
Paper: arxiv.org/pdf/2505.13100
Code (PyTorch/Captum/TensorFlow): github.com/esl-epfl/cro...
Many signals are better understood as frequencies, independent sources, trends, or seasonal components — not individual timestamps.
Cross-domain integrated gradients generates saliency maps directly in these domains, without modifying the original model.
Cross-domain integrated gradients generates saliency maps directly in these domains, without modifying the original model.
May 19, 2026 at 9:27 AM
Many signals are better understood as frequencies, independent sources, trends, or seasonal components — not individual timestamps.
Cross-domain integrated gradients generates saliency maps directly in these domains, without modifying the original model.
Cross-domain integrated gradients generates saliency maps directly in these domains, without modifying the original model.