SShortSingh.
Back to feed

Energy-based reasoner adds built-in progress signal to looped reasoning models

0
·5 views

Looped reasoning models reuse a small network repeatedly to gain depth without extra parameters, but they lack a way to judge their latent states, decide when to stop, or pick the best candidate answer. A paper from Duke, UIUC and Harvard proposes Energy-Based Reasoners, which learn an energy function that scores how compatible a latent state is with the input. The same function drives the update as gradient descent, so the model halts when the gradient becomes small and can select the lowest-energy trajectory among several. The authors report that their 1.8M-parameter model reaches 99.9% accuracy on Sudoku-Extreme using 345 times less inference compute than the next-best compared model. Training includes an exploration component meant to shape the energy landscape beyond the correct solutions and to handle puzzles with multiple valid answers.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Ch.05 Examines How Manipulator Morphology Shapes Robotic Capability

This chapter from the STATERA Library's Embodiment and Morphology volume explains manipulator morphology as the physical layout of links, joints, actuators, and end effectors in robots that interact with objects. It argues that a robot's mechanical structure shapes its perception, planning, learning, and control, so manipulation ability depends on designing the body and its computation together. The text outlines how joint arrangement determines reach, orientation, and leverage, and classifies manipulators as serial, parallel, or hybrid. It also covers degrees of freedom, including how redundancy aids obstacle avoidance and dexterity, and how workspace is defined by link lengths, joint limits, and mounting geometry.

0
ProgrammingDEV Community ·

Why Humanoid Robots Are Built Like Humans: Embodiment and Environment Design

This chapter from the STATERA Library's Volume 03 examines why robotics often adopts human-like morphology. The core argument is that human-built environments, including buildings, tools, and workspaces, are organized around human body dimensions, so a robot with a similar structure can operate in them without major infrastructure changes. The author stresses that the main benefit is functional compatibility rather than visual resemblance. Bipedal locomotion is presented as useful for stairs and uneven terrain, though it brings significant balance and energy-cost challenges compared with wheels. Human-like arms and hands are also highlighted for interacting with existing handles, switches, and tools, while the text notes that reproducing human dexterity remains difficult.

0
ProgrammingDEV Community ·

Pyrefly 1.5.0 Brings Speed Gains and Stricter Diagnostics That Could Break CI Pipelines

Pyrefly 1.5.0, the type checker's release six months after its first stable version, reports about 40% faster runs and 9% lower memory use than v1.4.1 across 71 benchmarks. Those gains make it easier for teams to make type checking a required pull-request gate rather than an optional step. The release also reaches 100% conformance on the 146-file Python typing conformance suite, which required tighter symbol name resolution inside annotations. As a result, code that passed under v1.4.1 may now trigger new type errors. Several previously editor-only checks, such as unused imports and unreachable code, are now standard error kinds at hint severity, and teams whose configurations raise them above hint could see previously silent issues fail builds.

0
ProgrammingDEV Community ·

Quadruped Robot Morphology: How Body and Leg Design Shape Locomotion

This educational chapter examines the physical configuration of four-legged robots, covering the arrangement of their trunks, limbs, joints, actuators, and contact surfaces. It explains that these structural features determine how a quadruped supports weight, generates motion, and responds to external forces. The text notes that such robots can traverse terrain that challenges wheeled platforms, drawing inspiration from biological quadrupeds while adapting to engineering constraints like electric actuation and control complexity. It describes typical leg designs with multiple actuated joints, along with trade-offs involving stance width, leg length, and mass distribution. The chapter concludes by highlighting how placement of components such as motors and heavy elements affects stability and responsiveness.