AI & Machine Learning

Generates novel, structurally plausible protein sequences from small alignments using a training-free stochastic attention mechanism on a standard laptop.

Explicit identity framing is not necessary and may be inferior for low-data LoRA safety fine-tuning.

Gauge-equivariant neural operators enable discretization-invariant and geometry-consistent solving of complex PDEs.

Adaptive computation for multimodal LLMs drastically reduces compute waste on easy cases while focusing on hard ones.

BrainBench exposes a significant gap between LLM benchmark performance and genuine commonsense reasoning.

A training-free operator for streaming 3D reconstruction reduces geometric drift using Grassmannian manifolds.

POLCA uses LLMs as stochastic optimizers with theoretical convergence guarantees for complex system-level tasks.

DynaAvatar achieves zero-shot 3D human reconstruction from a single image with motion-dependent cloth dynamics.

HO-SFL enables backprop-free fine-tuning on edge devices without the convergence penalty typical of zeroth-order methods.

Agent architectures require an explicit epistemic control layer to route questions between incompatible reasoning frameworks.

RAZOR provides a lightweight, targeted unlearning framework for Transformers and Diffusion models without retraining.

Demonstrates that safety and utility in LVLMs are not inherently antagonistic and can be simultaneously improved through inference-time projection.

Provides the first theoretical proof that dataset distillation efficiently encodes the low-dimensional structure of non-linear tasks.

Proves a fundamental expressivity limit where Message-Passing Graph Neural Networks are infinitely weaker than standard Color Refinement algorithms.

Introduces an asynchronous Mixture-of-Transformers architecture for autonomous driving that decouples slow reasoning from fast action execution.

Releases an 11-billion example dataset and model (RealVLG-R1) for unified real-world visual-language grounding and robotic manipulation.

Achieves over 80% of full-resolution VLM performance while using only 1% of the original pixel budget through bio-inspired foveated sampling.

Applies Signal Detection Theory to reveal that standard LLM calibration metrics conflate sensitivity (knowledge) with bias (confidence), leading to misleading evaluations.

A unified graph propagation library achieving 35,000x speedups, enabling full simulations on billion-edge graphs in seconds.

Releases a million-scale human preference dataset (29M pairs) specifically for text-to-image editing tasks.

Introduces 'Directional Routing', a lightweight mechanism that becomes the dominant computational pathway and enables transformers to self-organize into syntactic and adaptive regimes.

Recasts the LLM itself as a graph-native aggregation operator (Graph Kernel) for message passing on text-rich graphs.

Attention Residuals replace fixed-weight residual connections with softmax attention over preceding layers to prevent hidden-state dilution in deep LLMs.

MUNKEY introduces a 'design-to-forget' paradigm where machine unlearning is achieved through zero-shot key deletion rather than expensive parameter updates.

AdaAnchor enables LLMs to perform multi-step reasoning entirely in latent space with an adaptive halting mechanism to optimize compute.

AnoleVLA replaces the standard Transformer backbone in robotic Vision-Language-Action models with Deep State Space Models for a 3x speedup.

This paper reveals that pre-trained image editing models can be repurposed for video frame interpolation using only a few hundred LoRA samples.

Researchers identify 'Agentic Pressure' as a phenomenon where increased reasoning capability actually helps models rationalize and execute safety violations.

Writer-R1-4B outperforms 100B+ parameter models in creative writing by utilizing memory-augmented self-reflection and fine-grained criteria generation.

Euler Characteristic Surfaces achieve 98% accuracy on time-series classification with O(n) complexity, crushing previous topological methods that only hit 62%.

Small models (<=4B) fail document extraction not because of poor vision, but due to 'schema echo' where they copy the output structure instead of extracting data.

Ultra-low-bitrate image compression achieves 50% bitrate savings by treating decoding as a 'next-frame' video prediction task using diffusion priors.

Waypoint Diffusion Transformers (WiT) untangle pixel-space generation by using semantic waypoints, bypassing the need for information-lossy latent autoencoders.

LLM-based judges are negatively correlated with actual future research impact, systematically overvaluing 'novel-sounding' ideas that never materialize.

ForceVLA2 introduces explicit force awareness and hybrid control to Vision-Language-Action models, enabling stable contact-rich manipulation.

Recurrent gradient transport is massively redundant: propagating through just 6% of paths recovers nearly all adaptation ability in online learning.

The anonymity of leaderboards like LM Arena can be compromised using Interpolated Preference Learning to identify target models based on stylistic signatures.

SCAN enables reliable sequential knowledge editing in LLMs for up to 3,000 edits without the catastrophic forgetting or model collapse seen in current methods.

This physics-informed VLM framework improves physics-grounded anomaly detection AUROC from 66.9% to 96.7%.

HapticVLA achieves tactile-aware robotic manipulation at 86.7% success rate without requiring any physical tactile sensors at inference time.

IConE enables stable self-supervised learning even at batch size 1, overcoming the memory bottlenecks of high-dimensional scientific and medical data.

FlashU is the first framework to accelerate unified multimodal models by exploiting the distinct neuron sets used for generation vs. understanding.

GVC1D achieves over 60% bitrate reduction in video compression by replacing standard 2D latent grids with compact 1D latent tokens.

Tagarela releases 8,972 hours of high-quality Portuguese podcast audio, rivaling the scale of GigaSpeech for English.

MeMix is a training-free, plug-and-play module that reduces 3D reconstruction error by up to 40% in long sequences by mitigating state drift.

FuXiWeather2 is a unified end-to-end neural framework for weather assimilation and forecasting that outperforms global operational systems.

This paper proves that increasing test-time compute via beam search can actually hurt LLM reasoning performance due to overestimation bias.

Sparsity (MoE and GQA) is found to act as a critical regulator for variance propagation, mitigating the 'curse of depth' in LLMs.

Test-time reinforcement learning (TTRL) is found to amplify model harmfulness and jailbreak vulnerability when exposed to malicious prompt injections.

A large-scale study reveals that 78% of AI failures are 'invisible,' where the system fails without the user realizing or indicating an error.