ZeroPoint: Energy-Based Decision Model Makes Abstention a Zero-Energy State
kalyan_kpl · x · 2026-11-11
A new arXiv paper, ZeroPoint by Javier Marín, proposes a lightweight energy-based decision model (EBM).
- Problem: Standard EBMs only learn relative compatibility among options—adding the same input-dependent constant to all energies leaves predictions, the Gibbs distribution, and usual losses unchanged—so there is no learned reference level to decide that no option is acceptable.
- Method: ZeroPoint makes the empty answer a state with energy fixed at zero for every input, giving an explicit reference; each option's non-negative depth measures its compatibility relative to abstaining. Selecting an option vs. abstaining becomes part of ordinary inference trained with negative log-likelihood, replacing separate rejection rules.
- Implementation: The energy function is a small multilingual encoder reading the input with all options, invariant to their order. Ablations show most of the benefit comes from this training scheme.
More from Research
- Scandinavian Journal of Statistics editors reflect on publishing in the age of generative AI — michaelchchoi · 2026-11-11
- Claude Opus 5.5 runs an architecture search and finds AttnRes-Canon, a new top Transformer design — SonglinYang4 · 2026-11-11
- 5 research habits worth knowing: read papers backwards, reproduce first, wire coding agents via MCP — cneuralnetwork · 2026-11-11
- Running Qwen2.5-Coder-3B on Android: Jupyter + Pydantic Evals on-device eval pipeline — alex_b3b · 2026-11-11
- cloneofsimo releases a video explaining determinantal point processes — cloneofsimo · 2026-11-11
- Mapping Qwen3-8B's internal touch body map into an anatomical plate — repligate · 2026-11-11