Follow the latest research
alphaXiv connects papers, researchers, and organizations, grounding its answers in the underlying work.
A recurrent decoder carries computation across every prompt and response token, while exact full-history replay keeps reinforcement-learning policy states aligned with current parameters.
SJTU
Tsinghua
The survey’s five-level framework distinguishes persistent learning from genuine recursive improvement, clarifying which decisions AI systems control and which remain human-governed.
EPFL
KAIST
Training recurrent denoisers on progressively cleaner states enables reasoning models to improve with additional inference computation and generate diverse valid solutions.
Researchers to follow
View all
Are you a researcher? Find your profile
DeepseekCross-layer cache reuse, low-precision storage, and bounded replay make million-token multimodal agents substantially easier to serve within limited memory.
Stanford
UW
Repeated training data erodes Mixture-of-Experts language models faster than dense models, while dropout can partly restore generalization.
Pretraining teaches language models to counteract quantization errors across layers, while output geometry preserves their highest-confidence token predictions.
A shared pixel-space representation lets one model understand, reason about, generate, and edit images while preserving strong multimodal comprehension.
Westlake University
A frozen video world model can explore new camera paths while preserving an event’s appearance, timing, and previously generated scene states.
CMU
Multidimensional rarity measures expose culturally specific entities that popularity metrics miss, while reasoning-guided Wikipedia retrieval improves multilingual linking on this neglected long tail.
Incoming video streams can be edited in real time while preserving source motion and timing, enabling style transfer, virtual try-on, and scene replacement.
NTU
Grounding predicted state changes in temporally aligned robot experience helps manipulation policies adapt executable actions to new scenes and recover from disturbances.
Researchers to follow
View all
Repeated measurements in two qubit bases yield closed-form purification rates, while right-angle settings cannot reveal noise through this diagnostic.
Dartmouth College
OSU
Infrastructure-side evidence lets benchmarks distinguish merely exposed reward-hacking paths from exploits agents actually use during evaluation.
Muscle-driven reinforcement learning generates biomechanically realistic sprints and athletic drills without motion-capture demonstrations, enabling predictive studies of altered muscle properties.
Northwestern University
Microsoft Research
The benchmark reveals that multimodal models can recognize topological relations in static scenes but struggle to preserve them while planning actions.
The framework recasts proton PDF extraction as a coupled linear inverse problem, enabling model-agnostic reconstruction from diverse deep-inelastic-scattering measurements.
OpenAIA smooth, compactly forced three-dimensional flow can develop unbounded velocity in finite time while retaining uniformly bounded kinetic energy.
Yale
Brown University
Recursive latent-rollout training helps visual world models preserve dynamics-relevant information, enabling long-horizon prediction and robot control across unseen gravitational conditions.
HITRobots can use sparse visual memories to plan long-horizon manipulations while executing with fixed context, preserving historical evidence without growing inference inputs.
Shanghai AI Lab
Jointly predicting tokens and multi-token concepts enables language models to reach comparable training loss with substantially fewer tokens than standard architectures.
