AACL 2026 tutorial maps the diffusion language model field: non-autoregressive generation from foundations to frontiers
Bollegala · x · 2026-11-10
The closing tutorial of AACL-IJCNLP 2026, "Diffusion Language Models" by Daisuke Oba (Institute of Science Tokyo) and Danushka Bollegala (University of Liverpool / Amazon), targets NLP researchers with no diffusion background; slides and reading list are already available.
- Core idea: DLMs replace left-to-right decoding with iterative denoising — one model call can decide several tokens and each prediction sees bidirectional context; DLMs have reached LLM scale with open-weight models and commercial APIs
- Technical arc: defining diffusion for text, relating it to autoregressive and masked language modeling, surveying how large DLMs are built, and rethinking inference, training, and post-training (including RL) around the new generation process
- Open frontiers: evaluating reliability, safety, and social bias of DLMs; moving beyond discrete diffusion to continuous flow-based language models
- Multilinguality, multimodality, interpretability, diversity, and safety evals are flagged as wide-open opportunities for the NLP community, with demonstrated results separated from open problems
More from Research
- Four Free Math Books for ML: Linear Algebra, Probability, Optimization, Information Theory — cneuralnetwork · 2026-11-10
- 4 free math books that actually matter for ML: linear algebra to information theory — cneuralnetwork · 2026-11-10
- Meta FAIR's e-means beats k-means in 1/10 epochs with 10-45x GPU speedups — burny_tech · 2026-11-10
- NeurIPS workshop paper: gradient-based data attribution can't reliably filter subliminal learning — ponguru · 2026-11-10
- Yang-Mills Working Group to Release OpenAI's Full Lean Code and Validation Tomorrow — burny_tech · 2026-11-10
- Interpretability researchers reminded: explanation matters, not just predictive power — burny_tech · 2026-11-10