PAT3D: Physics‑Augmented Text‑to‑3D Generates Physically Stable Scenes

PAT3D, a physics‑augmented text‑to‑3D scene generation framework presented at ICLR 2026, extracts object relationships, initializes a hierarchical layout, and optimizes it via differentiable rigid‑body simulation, producing physically stable, semantically faithful scenes that can be directly edited, animated, or used for robot simulation, outperforming prior methods.

AIICLR2026physics simulation

0 likes · 9 min read

PAT3D: Physics‑Augmented Text‑to‑3D Generates Physically Stable Scenes

Machine Heart

Apr 4, 2026 · Artificial Intelligence

Does Scale Stealthily Hijack Attention? PMDformer’s Simple Subtraction Fix for Long-Term Forecasting

The paper identifies scale differences between patches as a hidden source of attention distortion in long‑term time‑series forecasting, introduces PMDformer with Patch Mean Decoupling, Neighbor Variable Attention, and Trend Recovery Attention, and demonstrates state‑of‑the‑art accuracy and efficiency across eight benchmark datasets.

ICLR2026Long-term Time Series ForecastingMultivariate Forecasting

0 likes · 8 min read

Does Scale Stealthily Hijack Attention? PMDformer’s Simple Subtraction Fix for Long-Term Forecasting

Machine Heart

Apr 1, 2026 · Artificial Intelligence

MeanCache Sets New Multi‑Modal Generation Inference Speed Benchmark at ICLR 2026

MeanCache introduces an average‑velocity‑driven caching framework that uses Jacobian‑vector‑product correction and a multigraph‑based scheduling algorithm to achieve over 4× speedup on state‑of‑the‑art multimodal diffusion models while preserving image fidelity and semantic consistency.

Average Velocity CachingDiffusion Model AccelerationICLR2026

0 likes · 5 min read

MeanCache Sets New Multi‑Modal Generation Inference Speed Benchmark at ICLR 2026

vivo Internet Technology

Mar 18, 2026 · Artificial Intelligence

How Ada-RefSR Eliminates Hallucinations in Single‑Step Diffusion Super‑Resolution

This article presents Ada-RefSR, a novel single‑step diffusion‑based reference super‑resolution framework that introduces a "Trust but Verify" paradigm, adaptive implicit correlation gating, and lightweight architecture to robustly suppress hallucinations and achieve state‑of‑the‑art performance on multiple benchmarks, while being suitable for mobile deployment.

Ada-RefSRICLR2026diffusion

0 likes · 10 min read

How Ada-RefSR Eliminates Hallucinations in Single‑Step Diffusion Super‑Resolution

Machine Learning Algorithms & Natural Language Processing

Feb 26, 2026 · Artificial Intelligence

LightMem (ICLR 2026): Cutting the Cost of Long‑Term Memory for Large Language Models

LightMem proposes a three‑stage, human‑inspired memory pipeline that dramatically lowers token usage, API calls, and latency while preserving accuracy, achieving up to 7.7% higher scores and 30‑plus‑fold cost reductions on long‑context benchmarks.

ICLR2026LLMLightMem