September 2026
Conference · Oral
A Projection-Based Surrogate Gradient Interpretation for Neural Codec Wrappers
IEEE MMSP 2026 · Istanbul, Türkiye
Reinterpreting the SCALED surrogate gradient as a first-order local approximation of the codec, and extending it from downscaling-only to full neural wrapper optimization, with BD-Rate (PSNR) reductions up to -23.59% on x264.
December 2025
Conference · Poster
SCALED: Surrogate-gradient for Codec-Aware Learning of Downsampling in ABR Streaming
PCS 2025 -Picture Coding Symposium · Aachen, Germany
Codec-aware learning framework enabling end-to-end training of neural downscalers through non-differentiable video codecs via surrogate gradients. 6.11% BD-Rate improvement over codec-agnostic approaches.