Transformers can store facts efficiently using associative memories.
problem Understanding how transformers store and recall factual information.
method Proved linear scaling of storage capacities for linear and MLP associative memories, introduced a synthetic task, and analyzed gradient flow.
result Shallow transformers can achieve near optimal storage capacity for factual recall tasks using associative memories.
New findings show tool-augmented models can recall unlimited facts, outperforming purely memorized models.
problem Limitations of purely memorized models in recalling large amounts of factual information.
method Demonstrated the benefits of in-tool learning (external retrieval) over in-weight learning (memorization) for factual recall.
result Proved that tool-use enables unbounded factual recall via a simple and efficient circuit construction.
LLMs can be tricked into recalling facts based on context clues.
problem Manipulation of LLMs' factual recall through context changes.
method Mathematical exploration of transformers' associative memory properties.
result Transformers use self-attention and value matrix for associative memory.
Sharp limits found for storing and retrieving input-output associations in linear associative memories.
problem Understanding the fundamental limits of storing and retrieving input-output associations in neural networks.
method Study of a minimal linear associative memory model, introducing a decoupled model and using statistical physics to characterize storage capacity.
result Linear associative memory can store up to 1/2 log(p) associations, providing a sharp statistical-physics characterization.
Transformers learn to recall with non-orthogonal embeddings in realistic settings.
problem Understanding how transformers store and retrieve knowledge in practical scenarios.
method Analyzing a single-layer transformer with random embeddings trained on a token-retrieval task.
result Explicit formulas for the model's storage capacity reveal a multiplicative dependence on sample size, embedding dimension, and sequence length.
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.
SLED improves factuality in LLMs without external knowledge.
problem Unreliable or factually incorrect outputs from large language models.
method Contrasts final layer logits with early layers' logits, uses approximate gradient to refine outputs.
result Consistently improves factual accuracy over existing methods.
MACI improves LLM factuality inference with higher retention and lower time cost.
problem Ensuring factuality in LLM responses for high-stakes domains.
method Reformulated conformal inference in a multiplicative filtering setting, leveraging ensembles for more accurate factuality scores and group-conditional calibration.
result MACI achieves higher retention and lower time cost compared to baselines, preserving validity through group-conditional calibration.
New methods for explaining Random Forest predictions using case-based reasoning.
problem Lack of explainability for black-box machine learning models like Random Forests.
method Extracting distance metric from Random Forests to identify prototypes, critics, counter-factuals, and semi-factuals.
result Identified special points from training datasets to explain Random Forest predictions.
FactTest assesses LLM factuality with Type I error control.
problem Lack of rigorous factuality verification for LLMs.
method Formulates factuality testing as hypothesis testing, ensuring Type I and II error control.
result Improves model accuracy by over 40% in abstaining from unknown questions.
Proposes a deep learning framework for estimating counterfactual outcomes.
problem Challenges in estimating individual outcomes under different treatments.
method Deep variational Bayesian framework integrating factual and similar subjects' outcomes.
result Rigorously integrates individual features and similar subjects' responses for counterfactual outcomes.
MGM improves media profiling by integrating textual and structural features.
problem Evaluating the bias and factuality of news media from digital data.
method MediaGraphMind (MGM) framework using variational EM, integrating textual and structural features.
result Achieves new state-of-the-art results in media profiling.
PO-Flow models potential and counterfactual outcomes for personalized treatment decisions.
problem Predicting individualized treatment effects from observational data.
method Continuous normalizing flow (CNF) framework for causal inference.
result Unified approach to potential outcome prediction, treatment effect estimation, and counterfactual prediction.
Graph signal processing detects hallucinations in large language models.
problem Detecting factual reasoning from hallucinations in large language models.
method Modeling transformer layers as dynamic graphs, using spectral analysis to define diagnostics.
result Spectral signatures can distinguish different types of hallucinations and achieve high accuracy.
Bounds on factual and counterfactual distributions under measurement error in discrete models.
problem Measurement errors in discrete data and their impact on inference.
method Expressing modeling assumptions as linear constraints and using linear programming to derive bounds.
result Sharp bounds on factual and counterfactual distributions for various models, including instrumental variable scenarios.
IGSD separates task-specific content channels in transformer components by comparing activation replacement with zero ablation.
problem Mechanistic interpretability of transformer components
method IGSD: paired-intervention framework for comparing activation replacement with zero ablation
result IGSD identifies an early-layer content channel in transformer components that standard importance methods underestimate.
We present SemEval-2019 Task 8 on Fact Checking in Community Question Answering Forums, which features two subtasks. Subtask A is about deciding whether a question asks for factual information vs. an opinion/advice vs. just socializing. Subtask B asks to predict whether an answer to a factual question is true, false or…
Estimates counterfactual outcomes linking observed and unobserved data.
problem Estimating expected counterfactual outcomes for individuals.
method Introduces retrospective counterfactual estimators and prediction intervals linking observed and unobserved outcomes.
result Retrospective counterfactual estimators and prediction intervals asymptotically satisfy valid coverage under standard causal assumptions.
This study quantifies uncertainty in comparing treatments using RCTs with before-and-after measures.
problem Uncertainty in comparing treatments using RCTs with before-and-after measures.
method New statistical modeling principle called ETZ enables counterfactual uncertainty quantification (CUQ) in RCTs with Before-and-After Repeated Measures.
result CUQ typically has lower variability than factual uncertainty quantification and can be achieved in RCTs.
We present a study on predicting the factuality of reporting and bias of news media. While previous work has focused on studying the veracity of claims or documents, here we are interested in characterizing entire news media. These are under-studied but arguably important research problems, both in their own right and …
Muon spectral optimizer outperforms SGD in associative memory tasks.
problem Understanding the advantage of spectral optimizers in learning associative memory.
method Linear associative memory problem, Gaussian inputs and outputs, power law frequency distribution, thresholded gradient approximation.
result Muon significantly outperforms SGD in storage capacity and recovery rates.
New algorithm reduces online learning regret for bounded recall games.
problem Reducing regret in online learning with limited past information.
method Constructing a stationary bounded-recall algorithm with O ( 1 / M ) O(1/\sqrt{M}) O ( 1/ M ) regret. result Any low regret bounded-recall algorithm must be aware of past losses' order.
This paper introduces a new method to train normalizing flows using precision-recall divergences.
problem Training generative models with mode dropping and low-quality samples.
method Introduces PR-divergences and proposes a novel generative model to minimize precision-recall trade-offs.
result Normalizing flows can be trained to achieve specific precision-recall trade-offs using PR-divergences.
Knowledge bases (KBs) are the backbone of many ubiquitous applications and are thus required to exhibit high precision. However, for KBs that store subjective attributes of entities, e.g., whether a movie is "kid friendly", simply estimating precision is complicated by the inherent ambiguity in measuring subjective phe…
The study examines how class imbalance affects precision-recall curves.
problem Understanding how precision changes with class imbalance ratios.
method Analyzes the relationship between precision, class imbalance ratio, and true/false positive rates.
result Predicts changes in precision-recall curves and other measures with class imbalance ratios.
Machine learning improves learning and memory retention by optimizing study sessions.
problem Improving learning and memory retention methods for factual material.
method Large-scale randomized controlled trial with machine learning optimization of study sessions.
result Study sessions optimized with machine learning lead to 67% longer retention and 50% higher return rate.
Causal inference analysis is the estimation of the effects of actions on outcomes. In the context of healthcare data this means estimating the outcome of counter-factual treatments (i.e. including treatments that were not observed) on a patient's outcome. Compared to classic machine learning methods, evaluation and val…
Prognosticator improves performance in non-stationary MDPs.
problem Real-world applications often violate the stationary assumption in reinforcement learning.
method A policy gradient algorithm that maximizes a forecast of future performance.
result Prognosticator outperforms online adaptation techniques in non-stationary MDPs.
CVIB uses information theory to learn counterfactuals from MNAR data without RCTs.
problem Debiasing learning from missing-not-at-random (MNAR) data in recommendation systems.
method CVIB, a variational information bottleneck, separates task-aware mutual information into factual and counterfactual parts.
result CVIB significantly enhances both shallow and deep models in recommendation systems.
The study formalizes temporal precision and recall for anomaly detection in sequences.
problem Insufficient understanding of precision and recall in sequential anomaly detection.
method Formalized temporal precision and recall measures, developed time-tolerant confusion matrices, and demonstrated statistical significance.
result Precision and recall may overestimate performance with temporal tolerance.
We present FAKTA which is a unified framework that integrates various components of a fact checking process: document retrieval from media sources with various types of reliability, stance detection of documents with respect to given claims, evidence extraction, and linguistic analysis. FAKTA predicts the factuality of…
Unified and extended precision-recall metrics for generative models.
problem Evaluation of generative models using scalar metrics like FID and IS.
method Unified various precision-recall approaches under a common framework.
result Recovered entire precision-recall curves and exposed metric pitfalls.
For information retrieval and binary classification, we show that precision at the top (or precision at k) and recall at the top (or recall at k) are maximised by thresholding the posterior probability of the positive class. This finding is a consequence of a result on constrained minimisation of the cost-sensitive exp…
Proposes a game-theoretic approach for class-dependent rationalization.
problem Optimizing feature selection for complex neural predictors.
method A game-theoretic approach where classes compete to find evidence for factual and counterfactual scenarios.
result The method identifies both factual and counterfactual rationales consistent with human rationalization.
This paper introduces DCE for better counterfactual explanations using optimal transport.
problem Lack of nuanced distributional characteristics in existing counterfactual explanations.
method Formulates a chance-constrained optimization problem using optimal transport to derive counterfactual distributions.
result DCE provides deeper insights into decision-making models by aligning counterfactual distributions with factual ones.
Recalls intrinsically harmonic forms and open problems.
problem Open problems related to intrinsically harmonic forms.
method Recalling definitions and known results.
result Clarification of intrinsically harmonic forms and open questions.
MetaCI uses meta-learning to generalize causal inference across heterogeneous populations.
problem Generalizing causal inference from multiple homogeneous subgroups to a heterogeneous population.
method MetaCI employs meta-learning to handle data distribution shifts and concept drifts.
result MetaCI outperforms random initialization and other methods in counterfactual inference.
Paper tackles imbalanced binary classification by optimizing precision and recall directly.
problem Imbalanced binary classification where standard accuracy is misleading.
method Exact constrained reformulations for precision and recall optimization.
result ERO framework outperforms state-of-the-art methods on multiple datasets.
LLMs can memorize economic data and recall exact values before their training cutoff.
problem Evaluating the trustworthiness of LLMs' economic forecasts during their training period.
method Demonstrated through counterfactual forecasting and analysis of LLMs' recall ability.
result LLMs have memorized economic and financial data, leading to recall-level accuracy before their knowledge cutoff.
HabitatAgent offers a multi-agent system for transparent housing consultation.
problem Opaque reasoning and brittle multi-constraint handling in housing recommendation systems.
method HabitatAgent is a multi-agent architecture with specialized roles for memory, retrieval, generation, and validation.
result HabitatAgent achieves 95% accuracy in real user consultation scenarios, significantly outperforming a strong baseline.
LLMs compress financial texts, but distort decision-making.
problem LLMs compress financial texts, altering decision-making.
method Analyzed two diagnostic patterns: decontextualization and model dependency. Proposed Agentic Context Compression.
result LLM-compressed financial texts alter decision-making.
Transformers can be hijacked by context, but deeper models are more robust.
problem Robustness of Transformers against context hijacking for linear classification.
method Developed a theoretical analysis on the robustness of linear transformers, considering model depth, training context lengths, and number of hijacking context tokens.
result Deeper transformers are more robust to context hijacking.
In this article we revisit the definition of Precision-Recall (PR) curves for generative models proposed by Sajjadi et al. (arXiv:1806.00035). Rather than providing a scalar for generative quality, PR curves distinguish mode-collapse (poor recall) and bad quality (poor precision). We first generalize their formulation …
RAGuard improves safety in LLMs for offshore wind maintenance.
problem Conventional LLMs fail with specialised or unexpected scenarios in offshore wind maintenance.
method Integrates safety-critical documents alongside technical manuals in RAG framework.
result RAGuard increases safety recall from almost 0% to over 50% while maintaining technical recall above 60%.
Paper proposes a human-algorithm approach to reduce medical device recall risk and workload.
problem High recall rate and regulatory workload in FDA's 510(k) pathway.
method Developed machine learning models to estimate recall risk and proposed a data-driven clearance policy.
result Conservative evaluation of policy shows a 32.9% improvement in recall rate and 40.5% reduction in workload.
Framework for precise recall control in spatial conflation tasks.
problem Precise recall control in large-scale spatial conflation tasks to avoid downstream analytics failures and excessive manual review.
method End-to-end framework using equigrid bounding-box filter, CSR representation, neural ranker, and inverse-variance weighted ensemble of threshold estimators.
result Achieves exact recall with sub-percent variance over tens of millions of geometry pairs, runs on a single TPU v3 core.
A new method for generating replay samples on the fly, optimizing for not forgetting.
problem Addressing the issue of forgetting in neural networks.
method Generates auxiliary samples on the fly using the model's implicit memory, specialized to each real training batch.
result Optimizing for not forgetting leads to more efficient and scalable generation of specialized samples.
Proposes a new tree-based algorithm for class-imbalanced data.
problem Classification of imbalanced data.
method Precision-Recall Curve (PRC) classification tree algorithm.
result The PRC classification tree outperforms classic algorithms for class-imbalanced data.