LaTRO optimizes latent reasoning in LLMs without external reward.
problem Training LLMs to perform complex reasoning tasks.
method Formulates reasoning as latent distribution sampling and optimizes via variational approaches.
result LLMs improve reasoning and evaluation quality through self-improvement.
A framework isolates VQA reasoning from perception for better model evaluation.
problem Improper separation of visual perception and reasoning in VQA models.
method Introducing a framework and a top-down calibration technique to decouple reasoning from perception.
result Improved evaluation of VQA models by separating reasoning from perception.
Forward-prediction models enhance physical reasoning, but only for specific tasks.
problem Improving physical reasoning in complex tasks involving many objects.
method Incorporated forward-prediction models into simple physical-reasoning agents and evaluated their performance on the PHYRE benchmark.
result Forward-prediction models improve physical-reasoning performance, especially on complex tasks, but generalization to new task templates is challenging.
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.
Self-supervised skip-tree training improves mathematical reasoning in language models.
problem Improving logical reasoning in language models for formal mathematics.
method Self-supervised language modeling on mathematical formulas, skip-tree task.
result Models trained on skip-tree task outperform standard models in mathematical reasoning tasks.
FinTradeBench benchmarks LLMs for financial reasoning combining company fundamentals and market signals.
problem Challenges in evaluating financial reasoning models for LLMs.
method Developed a benchmark integrating company fundamentals and trading signals, using a calibration-then-scaling framework.
result Clear performance gap between LLMs, retrieval improves reasoning over textual fundamentals but not trading signals.
Without relevant human priors, neural networks may learn uninterpretable features. We propose Dynamics of Attention for Focus Transition (DAFT) as a human prior for machine reasoning. DAFT is a novel method that regularizes attention-based reasoning by modelling it as a continuous dynamical system using neural ordinary…
This study explains RL training dynamics in LLMs, focusing on token-level optimization and reasoning pattern reshaping.
problem Understanding the training dynamics of RL in LLMs to improve their reasoning capabilities.
method Empirical analysis and theoretical modeling of RL training process, focusing on reasoning patterns and token optimization.
result RL primarily optimizes a sparse subset of critical tokens, reshaping reasoning pattern distributions and affecting model performance.
The paper studies how search and distillation improve reasoning in large language models.
problem Improving reasoning capabilities of large language models.
method Viewing chain-of-thought generation as a metastable Markov process, proving benefits of search and distillation.
result Search protocol rewards sparse edges, reducing the expected number of steps to reach different clusters.
Enhances large language models' reasoning through simpler off-policy reinforcement learning.
problem Improving large language models' ability to reason and solve problems.
method EM Policy Gradient, optimizing expected return over reasoning trajectories using Expectation-Maximization (EM) optimization.
result Achieves comparable or slightly superior performance to state-of-the-art methods on reasoning datasets, with additional cognitive behaviors.
Inferring new facts from existing knowledge graphs (KG) with explainable reasoning processes is a significant problem and has received much attention recently. However, few studies have focused on relation types unseen in the original KG, given only one or a few instances for training. To bridge this gap, we propose Co…
A new method for math reasoning that allows for iterative correction.
problem Standard reasoning models commit to each token and cannot recover from early errors.
method Generative framework with latent thought vectors for iterative self-correction.
result 30 rethinking iterations surpass baselines with 15 times more parameters.
VTA combines verbal and latent reasoning for accurate stock time-series forecasts.
problem Challenges in combining textual analysis with time-series data for financial forecasting.
method Converts stock price data into textual annotations, optimizes reasoning trace using inverse MSE, conditions time-series model outputs on reasoning attributes.
result VTA achieves state-of-the-art forecasting accuracy and interpretable reasoning traces.
This work investigates how multi-round reasoning improves LLM performance.
problem Improving problem-solving abilities in complex tasks with LLMs.
method Investigates approximation, learnability, and generalization properties of multi-round auto-regressive models.
result Transformers with finite context windows are universal approximators for Turing-computable functions and can approximate any Turing-computable sequence-to-sequence function through multi-round reasoning.
SE-RRMs solve structured problems like Sudoku and ARC-AGI by enforcing symbol equivariance.
problem Structured reasoning problems like Sudoku and ARC-AGI.
method Symbol-equivariant recurrent reasoning models enforcing permutation equivariance.
result SE-RRMs outperform prior RRMs on 9x9 Sudoku and generalize to larger and smaller instances.
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.
Fractured Sampling improves LLM reasoning efficiency by truncating CoT trajectories.
problem Efficiently scaling reasoning in large language models with limited tokens.
method Integrating truncated Chain-of-Thought (CoT) with Fractured Sampling across multiple dimensions.
result Fractured Sampling achieves superior accuracy-cost trade-offs compared to full CoT.
Transformers learn multi-step reasoning through gradient descent.
problem Understanding how transformers solve symbolic multi-step reasoning tasks.
method Theoretical analysis of gradient descent dynamics and multi-phase training.
result Trained one-layer transformers can solve both backward and forward reasoning tasks with generalization guarantees.
Improved reasoning model by sampling from power distribution without additional training.
problem Efficiently sampling from a sharpened distribution to improve reasoning models.
method Entropy-Cut Metropolis-Hastings algorithm that identifies key decision points for resampling.
result The method consistently improves reasoning models across various datasets.
Transformers with CoT don't enhance reasoning power across all tasks.
problem Does CoT enhance the reasoning power of transformers?
method Examined the memorization capabilities of fixed-precision transformers with and without CoT.
result Transformers with CoT cannot memorize all reasoning tasks, leading to a negative answer.
Hybrid framework injects TSLM insights into GRLM for robust time-series reasoning.
problem Lack of domain-specific knowledge in large language models for time-series reasoning.
method Hybrid knowledge-injection framework combining RLVR for efficient knowledge transfer.
result Consistently outperforms existing models by 7.9%-26.1% on multivariate time-series benchmarks.
Few-shot visual reasoning model learns analogical relationships from small data.
problem Training deep models on few samples for visual reasoning tasks.
method Meta-analogical contrastive learning to enforce structural similarity between training and test samples.
result Method outperforms state-of-the-art on RAVEN dataset with scarce training data.
Benchmark for math reasoning models from human proofs.
problem Measuring and accelerating machine learning models in high-level mathematical reasoning.
method Built a non-synthetic dataset from theorem prover proofs, defined a task for model to fill in missing propositions, used hierarchical transformer to improve performance.
result Neural models can capture non-trivial mathematical reasoning, hierarchical transformer outperforms baseline.
Deep learning is very effective at jointly learning feature representations and classification models, especially when dealing with high dimensional input patterns. Probabilistic logic reasoning, on the other hand, is capable to take consistent and robust decisions in complex environments. The integration of deep learn…
The study compares feed-forward and attention layers in language models.
problem Understanding the role of feed-forward and attention layers in language models.
method Empirical and theoretical analysis in a synthetic setting.
result Feed-forward layers learn simple distributional associations, while attention layers focus on in-context reasoning.
This work identifies and mitigates reasoning shortcuts in Neuro-Symbolic models.
problem Neuro-Symbolic models can achieve high accuracy by using unintended concepts.
method Characterized reasoning shortcuts as unintended optima of the learning objective and identified four key conditions.
result Reasoning shortcuts are difficult to mitigate, casting doubt on NeSy solutions' trustworthiness and interpretability.
Auto-CEI improves LLM reasoning by balancing assertiveness and conservativeness.
problem Hallucinations and laziness in LLM reasoning tasks.
method Expert Iteration explores reasoning trajectories, guiding incorrect paths back on track and promoting appropriate 'I don't know' responses.
result Auto-CEI achieves superior alignment in logical reasoning, mathematics, and planning tasks.
Neural Logic Reasoning integrates deep learning and symbolic logic for better prediction tasks.
problem Lack of cognitive reasoning in deep neural networks limits their ability to solve complex prediction tasks.
method Proposes Logic-Integrated Neural Network (LINN) that learns logical operations and conducts propositional logical reasoning.
result LINN significantly outperforms state-of-the-art recommendation models in Top-K recommendation.
Achieving artificial visual reasoning - the ability to answer image-related questions which require a multi-step, high-level process - is an important step towards artificial general intelligence. This multi-modal task requires learning a question-dependent, structured reasoning process over images from language. Stand…
Theoretical work shows integrating coherent reasoning improves LLM performance and error correction.
problem Improving reasoning and error correction in large language models (LLMs) with few-shot prompting.
method Theoretical analysis and sensitivity experiments on transformer behavior with coherent reasoning and corrupted demonstrations.
result The transformer gains better error correction ability and more accurate predictions when coherent reasoning is integrated.
New method for LLMs to learn reasoning by optimizing latent variables.
problem Teaching LLMs to generate logical justifications for answers.
method Formalized reasoning as latent variable model, derived FEM objective, designed sampling schemes.
result Prompt Posterior Sampling (PPS) outperforms other schemes in learning to reason.
Large language models can't efficiently reason conditionally in a distribution-free setting.
problem Impossibility of conditional PAC-efficient reasoning in large language models.
method Proof of impossibility in a distribution-free setting for non-atomic input spaces.
result Any algorithm achieving conditional PAC efficiency must defer to the expert model with high probability.
Discovering and exploiting the causal structure in the environment is a crucial challenge for intelligent agents. Here we explore whether causal reasoning can emerge via meta-reinforcement learning. We train a recurrent network with model-free reinforcement learning to solve a range of problems that each contain causal…
Tab-TRM uses recursive model for insurance pricing on tabular data.
problem Insurance pricing on tabular data.
method Adapts recursive latent reasoning to insurance modeling using a compact, parameter-efficient network.
result Improves insurance pricing accuracy using iterative refinement of latent tokens.
Reasoning models generate differently based on problem difficulty, not just length.
problem Understanding how reasoning models handle different problem difficulties.
method Examined hidden-state trajectories across competitive programming, mathematics, and Boolean satisfiability.
result Corrected trajectory geometry shows difficulty-dependent differences in reasoning models, with stronger effects in the code domain.
Paper introduces scalable neural architecture for solving NP-hard problems.
problem Solving NP-hard reasoning problems from natural inputs.
method Scalable neural architecture and loss function for discrete Graphical Models.
result Empirically shows efficient learning of NP-hard problems.
Neurosymbolic predictors fail to model uncertainty under independence assumption.
problem Neurosymbolic predictors' reliance on independence assumption limits their ability to model uncertainty.
method Formal analysis of NeSy predictors under independence assumption.
result Assuming independence among symbolic concepts prevents NeSy predictors from representing uncertainty.
Trade-R1 bridges verifiable rewards to stochastic financial markets via process-level reasoning verification.
problem Extending RL to financial markets where rewards are verifiable but noisy.
method A verification method that transforms reasoning over financial documents into a structured RAG task, using a triangular consistency metric.
result DSR achieves superior cross-market generalization while maintaining reasoning consistency.
RACER optimizes LLM-as-judge accuracy with dynamic reasoning selection.
problem Balancing reasoning accuracy with computational cost in LLM-as-judge settings.
method Formulates routing as a constrained distributionally robust optimization problem, accounting for distribution shift via KL-divergence uncertainty set.
result RACER achieves superior accuracy-cost trade-offs under distribution shift.
SRN improves set representations for relational reasoning.
problem Set permutational invariance limitations in existing approaches.
method Proposed a Set Refiner Network (SRN) to respect set invariance.
result Substantial gains in prediction performance and robustness on relational reasoning tasks.
Proposes CLRS-Text, a new benchmark for evaluating LM reasoning capabilities.
problem Lack of transferable benchmarks for evaluating reasoning capabilities of language models.
method Developed a textual version of the CLRS benchmark, generating diverse algorithmic tasks.
result Demonstrates a novel challenge for the LM reasoning community and validates prior work.
G1 uses RL to enhance LLMs' graph reasoning, improving performance on diverse tasks.
problem Limited graph reasoning abilities of LLMs, especially in synthetic graph-theoretic tasks.
method Curated synthetic graph dataset, RL training on LLMs.
result Significant improvements in graph reasoning, zero-shot generalization to unseen tasks.
REALFIN benchmarks financial reasoning by removing implicit assumptions, revealing model weaknesses.
problem Models struggle when implicit assumptions are missing, leading to incorrect answers.
method Developed a bilingual benchmark that systematically removes essential premises from financial questions.
result General-purpose models over-commit, while finance-specialized models fail to identify missing premises.
Transformers improve logical reasoning on longer proofs but struggle with length.
problem Understanding systematic generalization in neural proof generation.
method Soft theorem-proving using Transformer models, evaluating logical consistency and inference accuracy.
result Transformers improve generalization with longer proofs but have difficulty with length.
ChatGPT improves financial reasoning, overcoming biases in gold investment.
problem Improving financial reasoning and overcoming biases in investment decisions.
method Applied advanced prompt engineering and semantic news information to enhance LLMs' performance.
result ChatGPT with CoT prompt provides more explainable predictions and higher investment returns.
The AI2 Reasoning Challenge (ARC), a new benchmark dataset for question answering (QA) has been recently released. ARC only contains natural science questions authored for human exams, which are hard to answer and require advanced logic reasoning. On the ARC Challenge Set, existing state-of-the-art QA systems fail to s…
Attention mechanisms have been boosting the performance of deep learning models on a wide range of applications, ranging from speech understanding to program induction. However, despite experiments from psychology which suggest that attention plays an essential role in visual reasoning, the full potential of attention …
Method guarantees coherent factuality for language model outputs in reasoning tasks.
problem Ensuring correctness of language model outputs in reasoning tasks.
method Developed a conformal-prediction-based method applied to subgraphs within a deducibility graph.
result Achieved coherent factuality across target coverage levels, 90% on stricter definition.