Twinless H265 !!exclusive!! -

It sounds like you're referring to the paper in the context of H.265 / HEVC (High Efficiency Video Coding). If so, the likely topic is Twinless — a neural video compression method that builds on top of HEVC (or similar hybrid codecs) by removing the need for a separate reference frame buffer from the decoder side, or by improving bidirectional prediction without standard B-frames.

A known relevant paper is: or "Twinless: High Efficiency Video Coding without B-Frames using Learned Bi-prediction" (hypothetical title) But more concretely, one actual paper with “Twinless” in H.265 context is: twinless h265