Four papers we submitted together with our joint research team have all been accepted to ICML 2026 (Seoul, Korea), the top-tier international conference on machine learning. The work spans efficient language generation, diffusion large-model inference, and multimodal generation.
Accepted Papers
- Jacobi Forcing — Fast and Accurate Causal Parallel Decoding using Jacobi Forcing, a next-generation algorithm for parallel decoding of autoregressive large models
- d3LLM — Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing, the first time diffusion language models surpass autoregressive models in inference speed
- LightningRL — Breaking the Accuracy–Parallelism Trade-off of Block-wise dLLMs via Reinforcement Learning
- Think-Then-Generate — Reasoning-Aware Text-to-Image Diffusion with LLM Encoders, injecting LLM reasoning into text-to-image diffusion models
These results will progressively land in Triverse AI’s image-generation and edge-inference product lines, further reducing on-device inference latency and improving the controllability of multimodal generation.