Showing papers from Microsoft Research Asia Show all papers
Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
Atlanta Poster Session 6, Fri, Dec 11, 4:30 PM–7:30 PM, Hall C1 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
Atlanta Poster Session 3, Thu, Dec 10, 10:00 AM–1:00 PM, Hall C1 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
DPIAgent divides reproduction test generation into isolated diagnosis and test phases with structured handoffs, achieving up to 86.17% success on SWT-Bench Verified.
Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
P2T uses reference patches as privileged supervision to curate shorter, grounded agent trajectories via bi-objective optimization, improving SWE-bench Pass@1 by up to 10.8 points with ~15% lower inference cost.
Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
XL-DocBench introduces a human-verified benchmark for extra-long professional document understanding spanning thousands of pages with multi-page evidence, showing current systems still struggle with long-context structured reasoning.
Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026 · ▲ 2 on Hugging Face
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
DocAtlas treats long-document understanding as a mutable-state interaction process via a document harness with search, memory, and review tools, reaching 71.4% on MMLongBench-Doc and boosting a 4B VLM to 63.7% via reinforcement learning.
Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 1 on Hugging Face · Code ★ 12
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
Model-guidance replaces classifier-free guidance by training on condition posterior probabilities, doubling inference speed and achieving 1.34 FID on ImageNet 256.
Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026 · ▲ 8 on Hugging Face · Code ★ 175
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
Memory Grafting uses frozen hidden states from a grafting model as conditional n-gram memory for language models, improving average benchmarks to 53.86 versus 52.43 for vanilla Engram at 2.8B scale with minimal overhead.
Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
G2PO transforms agent trajectories into state-transition graphs to reduce variance and improve credit assignment, outperforming GRPO by up to 22.2% on long-horizon benchmarks.
Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
Standardized item-level benchmark releases should become AI evaluation infrastructure because aggregate scores obscure validity failures; OpenEval archives 10M responses to enable auditability and recover benchmark validity evidence.
Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
OneVision-Encoder applies codec-aligned sparsity to video, processing only high-entropy regions to outperform dense backbones with fewer tokens. It achieves 4.1% higher video accuracy than Qwen3-ViT across 16 benchmarks.
Paris Poster Session 5, Fri, Dec 11, 11:30 AM–1:30 PM, Paris Poster Hall · Published 2026 · ▲ 52 on Hugging Face · Code ★ 403
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.
LCVN introduces a language-conditioned navigation benchmark and compares diffusion-based latent imagination against unified autoregressive prediction for embodied agents. Latent imagination yields more temporally coherent rollouts, while unified prediction generalizes better to unseen environments.
Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026
Readers and the AI panel: vote on this paper to see what they said.
Only vote on papers you've read. Sign in with GitHub to vote.