Research, distilled.
Research explained clearly, with lessons that connect the ideas.
Curvature-guided compression avoids rotating model weights
At two bits per weight, trellis codecs compress model weights, but leading approaches rotate them first and must undo that transform during generation. Can curvature information guide the search directly, without rotation, while retaining accuracy on language tasks? CurveTQ is a rotation-free codec that weights its search by curvature. On three tested 4–8B Instruct models, it scored higher on average than two competing codecs after they received the same stored block-start information.
Latest added:
Latest lessonRoboticsWhen a Robot Struggles, What Should Its Training Change?
A robot may close a laptop yet drop an object it has grasped; another must feel whether an object is soft, or complete a task when the tabletop changes. Improving these behaviors is not one problem with one remedy. The studies here change different parts of a robot system: training from human touch, rewarding distinct successful actions during practice, or turning human movement into commands a humanoid can use. Each reports a physical-robot outcome, but the tasks and comparisons differ. The useful question is which change addresses the difficulty a particular deployment presents.
Latest added:
Latest across topics
View all papersChoosing fewer, better tokens improves distillation
When a student learns from its own answers, should every token count equally? The token reward measures teacher-student disagreement but can give large weight to tokens both models consider unlikely. DIAL-OPD, a token-selection method, also weighs disagreement by probability and trains on a chosen fraction. Across four tested model pairs, keeping 40% of tokens raised mean accuracy by up to 5.25 percentage points over full-token Vanilla OPD on seven tests of math problem solving, where accuracy measures correct answers.
RoboticsGLIO2 completes a route with weak LiDAR geometry
LiDAR mapping can fail when a road or landscape has too little shape, while a mistaken scan match can become a confident pose estimate that satellite data cannot easily correct. How can a localization system keep revising local geometry while still finding its place globally? GLIO2 jointly updates LiDAR, inertial, and raw satellite-navigation measurements; on one long bridge with weak LiDAR geometry, it completed the route while the evaluated LiDAR-based systems diverged.
Software & CodeChronos helps code agents search connected project history
A code change can depend on an old design decision that the current files do not explain. A search for similar descriptions may miss related earlier changes that describe a different problem. Chronos turns pull requests, or proposed code changes, into linked experience cards for agents to search. In evaluations, its full workflow solved more issue-resolution tasks than SWE-Agent with the same tested model, averaged across six models on one task set.
Trending this week
View all papersTop trending scores among papers added in the last seven days.
TopoGPU streams image topology calculations through GPUs
Persistent homology tracks how image structures appear and disappear across scales, but calculating it can slow GPU-based training when image data must move to the CPU. Can most of the work stay on the GPU while preserving the result? TopoGPU streams image chunks through GPU calculations, then reduces a matrix on the CPU. Across datasets, its end-to-end runs from loading through reduction averaged 53.24 times faster than CRipser, a CPU method for image cubical complexes, on an RTX 4080.
#1 · Trending 45.4RoboticsRobot history helps most when video learning predicts forward
A robot’s current image can show where an object is, yet miss whether it is moving or how an interaction is unfolding. Giving the controller more past images may help, but processing them can delay its next action. Long-WAM is a robot-control framework that predicts how a scene will change before generating actions. In RoboCasa GR-1, simulated tasks involving several object transfers, separately trained models completed more tasks with 19.2 seconds of history than with 2.4 seconds, with forecast and action lengths fixed.
#2 · Trending 45.2Software & CodeGit histories reveal how agent skills spread
A repository snapshot shows where an agent skill exists, but not who copied it or whether later fixes reached those copies. That makes it hard to trace a skill’s source or decide where review could interrupt its spread. Skill Constellations reconstructs dated copy links from Git histories and ranks repositories for review. In a later period, auditing its top 100 choices prevented a larger share of future adoptions of flagged high-risk skills than auditing the 100 most-starred repositories.
#3 · Trending 43.5Lessons
View all lessonsGuided paths that connect several papers into one idea you can keep.
When a Robot Struggles, What Should Its Training Change?
A robot may close a laptop yet drop an object it has grasped; another must feel whether an object is soft, or complete a task when the tabletop changes. Improving these behaviors is not one problem with one remedy. The studies here change different parts of a robot system: training from human touch, rewarding distinct successful actions during practice, or turning human movement into commands a humanoid can use. Each reports a physical-robot outcome, but the tasks and comparisons differ. The useful question is which change addresses the difficulty a particular deployment presents.
Machine LearningReaching the Goal, Responding in Time, and Staying Safe
A robot can complete a task and still respond too late to a moving object. It can also reach a destination after a collision. Saying that it “succeeded” leaves out what was measured and what happened while it decided. The selected studies offer a bounded answer: adapting computation was checked against task completion, rewarding varied successful behavior was checked under shifted simulation conditions, and a moving-world test separately counted safe arrivals. Each claim belongs to its own test.
Machine LearningWhat Else Changes When You Fix a Model?
A model can learn from a new corpus while losing old behavior, receive privacy-protecting noise while losing task accuracy, or refuse a harmful prompt while also refusing a harmless neighbor. Success at the intended target leaves a nearby question unanswered. The studies suggest a bounded way to investigate: locate what changes, choose a comparator, and inspect the closest relevant boundary. That boundary is different for training updates, privacy calibration, and prompt repair. Neither a plausible mechanism nor a reassuring broad score settles the other studies’ questions.