ResLRB: constant-memory differentiable light tracing via stochastic graph compression
ssh4net · x · 2026-10-10
A new paper, Stochastic Graph Compression for Constant-Memory Differentiable Light Tracing by Linas Beresna and Eugene Fiume, proposes ResLRB (Reservoir Light Replay Backpropagation).
It tackles reverse-mode differentiable light tracing memory blowup: naive autodiff records a computation graph that grows with the number of valid sensor connections per light path, while existing adjoint path replay doesn't extend to the splatting case. ResLRB compresses the adjoint graph during the primal pass by stochastically retaining a single representative sensor connection per path via a streaming weighted reservoir, making memory constant in both path length and sensor connections.
More from Research
- Xiaomi's MiMo-V2.6: agents run their own RL loop, DeepSWE score hits 72.6 for $2.6M — rohanpaul_ai · 2026-10-10
- New fastest deterministic 3SUM algorithm hits n^1.9961, matching randomized bound — basedjensen · 2026-10-10
- OpenAI's Navier-Stokes proof drama, explained: Lean version auto-verifiable, human-readable proof flawed — qi2peng2 · 2026-10-10
- Google paper frames LLM routing as Pandora's Box costly-search problem, with closed-form value-of-information policies — _onionesque · 2026-10-10
- QWM method: morphology encoder, adaptive reward normalizer, and latent morphology conditioning — breadli428 · 2026-10-10
- Factuality evals need a rethink: COLM workshop best paper argues current benchmarks are broken — caglarml · 2026-10-10