On-device research index

arXiv research

A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.

169,051 papers · 148 categories

Trend · papers per month

count % of corpus

5 results for “overthinking”

Shallow-Deep Networks: Understanding and Mitigating Network Overthinkingcs.LG

SDN mitigates overthinking in deep networks, reducing inference cost and preserving accuracy.

problem Overthinking in deep neural networks, leading to unnecessary computations and misclassifications.

method Introducing internal classifiers (Shallow-Deep Network) to SDNs, enabling early confidence-based exits.

result SDNs reduce inference cost by more than 50% and preserve accuracy, mitigating overthinking.

overthinking deep neural networks Shallow-Deep Network confidence-based early exits inference cost reduction

2018-10-16abs ↗pdf ↗

When More Sampling Hurts: The Modal Ceiling and Correlation Ceiling of Test-Time Scalingcs.LG

Sampling more can make models more confident in wrong answers, not better.

problem The modal ceiling and correlation ceiling limit the benefit of increased sampling.

method Analyzes the trade-offs between sampling more and selecting the best answer.

result Extra sampling beyond a certain point does not improve model performance and can even degrade it.

sampling test-time scaling modal ceiling correlation ceiling selection

2026-06-27abs ↗pdf ↗

Statistical Early Stopping for Reasoning Modelscs.AI

Early stopping methods reduce unnecessary reasoning steps in LLMs by monitoring uncertainty signals.

problem LLMs sometimes generate unnecessary reasoning steps, especially under uncertainty.

method Statistically principled early stopping methods that monitor uncertainty signals during generation.

result Uncertainty-aware early stopping improves efficiency and reliability in LLM reasoning, especially in math reasoning.

early stopping LLMs uncertainty renewal process math reasoning

2026-02-15abs ↗pdf ↗

EERO: Early Exit with Reject Option for Efficient Classification with limited budgetstat.ML

EERO optimizes resource usage for efficient classification.

problem Managing computational resources in complex machine learning models.

method EERO uses multiple classifiers with a reject option to adaptively shorten processing paths.

result EERO effectively manages budget allocation and enhances accuracy in complex scenarios.

machine learning resource management early exit reject option Bayesian risk

2024-02-06abs ↗pdf ↗

An Asymptotic Theory of Chain-of-Thought in In-Context Learningstat.ML

Study reveals how depth of reasoning affects generalization in models.

problem Understanding scaling behavior of generalization with CoT depth.

method Theoretical model of CoT in linear regression using random matrix theory.

result Sharp phase transition between exponential and polynomial improvement, saturation, and overthinking.

chain-of-thought in-context learning random matrix theory generalization phase transition

2026-06-02abs ↗pdf ↗