Improved unsupervised probing for ranking tasks using Contrast-Consistent Ranking.
problem Improving self-consistency in language model rankings.
method Adapting Contrast-Consistent Search (CCS) to Contrast-Consistent Ranking (CCR) for ranking tasks.
result CCR probing outperforms prompting techniques across different models and datasets.
Study examines flaws in probing LLMs' knowledge and introduces a new method.
problem Flaws in existing methods for probing the veracity of LLMs' internal knowledge.
method sAwMIL (Sparse-Aware Multiple-Instance Learning) combining multiple-instance learning with conformal prediction.
result LLMs encode a third type of signal distinct from true and false.
This paper enhances language models with knowledge awareness.
problem Understanding how much knowledge pretrained language models grasp.
method Inserting explicit knowledge layers into pretraining without changing transformer architecture.
result Significantly more knowledge packed into transformer parameters.
Study quantifies gender bias in language models across 7 languages.
problem Measuring gender bias in language models across multiple languages.
method Curated dataset of politicians, multilingual language models, probing language models.
result Larger language models do not show significant gender bias compared to smaller ones.
ZDP detects drift in large language models without labels, proving key theorems and metrics.
problem Detecting drift in large language models without task labels or output evaluations.
method Zero-Direction Probing (ZDP) framework based on null directions of transformer activations, proving theoretical guarantees.
result Proves the Variance--Leak Theorem, Fisher Null-Conservation, Rank--Leak bound, and logarithmic-regret guarantee.
DatedGPT prevents lookahead bias in financial forecasting models.
problem Lookahead bias in large language models trained on internet-scale data.
method Time-aware pretraining with annual data cutoffs and instruction fine-tuning.
result Models' knowledge is effectively bounded by their data cutoff year, improving forecasting validity.
The paper explores linear representations in language models using counterfactuals.
problem Understanding linear representations and geometric concepts in large language models.
method Formalized linear representation in output and input spaces, identified causal inner product.
result Unified understanding of linear representations and their connection to interpretation and control.
Language models learn automotive complaints, improving defect detection.
problem Detecting technical quality issues in customer feedback.
method Continual pre-training on ODI Complaints data set.
result Models achieve high precision in identifying technical terms.
Method reveals hidden token embeddings of large language models.
problem Understanding hidden token embeddings in large language models.
method Structured prompts to expose token input embeddings up to homeomorphism.
result Mathematical proof for generic LLMs shows effectiveness of method.
Benchmark tests spoken language models for infant language learning.
problem Understanding how infants learn language from speech.
method Developed a language-acquisition-friendly benchmark.
result Benchmarking shows models' strengths and weaknesses.
PROBE optimizes best-arm identification with cheap proxies, improving sample complexity.
problem Fixed-confidence best-arm identification with costly rewards and correlated cheap proxies.
method PROBE uses control-variate adjustment and phase elimination to learn residual variance online.
result PROBE achieves oracle sample complexity up to a constant factor and additive calibration cost.
Study quantifies how LLMs capture higher-order statistical structure using cumulant expansion.
problem Understanding how LLMs internalize statistical structure during next-token prediction.
method Cumulant-expansion framework treating softmax entropy as perturbation around center distribution.
result Cumulants reveal distinct signatures for mathematical vs. general text prompts, quantifying feature-learning dynamics.
Study bandwidth-limited training and inference of language models.
problem Training and inference of language models on scattered data with limited bandwidth.
method Analyzed two protocols: Federated Probe-Logit Distillation (FPLD) for training and Federated Conformal RAG (FC-RAG) for inference.
result Explicit high-probability KL-consistency rate and distribution-free marginal-coverage bound for Federated Conformal RAG.
New method tightens federated probe-logit distillation rates under varying bandwidths.
problem Estimating conditional distributions in federated learning with heterogeneous bandwidth constraints.
method Developed a new federated probe-logit distillation (FPLD) method with optimal allocation for varying bandwidths.
result Achieved matching lower and upper bounds for the minimax rate under heterogeneous bandwidths.
This two-part work puts forth the idea of engaging power electronics to probe an electric grid to infer non-metered loads. Probing can be accomplished by commanding inverters to perturb their power injections and record the induced voltage response. Once a probing setup is deemed topologically observable by the tests o…
Probe-level models have led to improved performance in microarray studies but the various sources of probe-level contamination are still poorly understood. Data-driven analysis of probe performance can be used to quantify the uncertainty in individual probes and to highlight the relative contribution of different noise…
New method predicts aphasia severity with narrower uncertainty intervals.
problem Predicting aphasia severity in stroke patients using neuroimages.
method Sparse heteroscedastic Bayesian high-dimensional regression with H-PROBE algorithm.
result H-PROBE provides narrower prediction intervals for aphasia severity.
New methods predict language model out-of-distribution behaviors using causal mechanisms.
problem Predicting how language models behave on unseen data.
method Two methods: counterfactual simulation and value probing.
result Both methods achieve high AUC-ROC and outperform causal-agnostic approaches in out-of-distribution settings.
A probing scheme is considered with an accessible and controllable qubit, used to probe an out-of equilibrium system consisting of a second qubit interacting with an environment. Quantum spontaneous synchronization between the probe and the system emerges in this model and, by tuning the probe frequency, can occur both…
Transformers encode latent distributions in text, improving performance in out-of-distribution cases.
problem What should embeddings from language models represent?
method Connecting autoregressive prediction to sufficient statistics, identifying three settings.
result Transformers encode latent generating distributions, improving performance.
We identify spectral conditions for reliable neural probe interpretation.
problem Unreliable performance of linear probes in interpreting neural representations.
method Formalized Spectral Identifiability Principle (SIP) based on eigengap and Fisher error.
result Reliability of neural probes depends on the eigengap relative to Fisher estimation error.
Optimal probing framework for scalable network monitoring.
problem Efficiently monitor growing cloud networks with limited budgets.
method A- and E-optimal experimental designs, Frank-Wolfe algorithm approximations.
result Significant reduction in probing budget with low estimation errors.
Paper introduces Manifold Probe for discovering representation manifolds in superposition.
problem Discovering representation manifolds in complex superposition representations.
method Generalizes linear regression probes to learn feature spaces and directions in superposition representations.
result Demonstrates Manifold Probe on Llama 2-7b representations, finding causally involved manifolds in model behaviour.
OLPA optimizes online user-centric selection with probing, achieving near-optimal regret bounds.
problem Sequential decision-making with unknown resources and rewards.
method Probing-augmented user-centric selection (PUCS) framework, greedy probing algorithm, OLPA algorithm.
result OLPA achieves a near-optimal regret bound of O ( T + ln 2 T ) \mathcal{O}(\sqrt{T} + \ln^{2} T) O ( T + ln 2 T ) for online settings. Study investigates how simple speech sounds can form abstract categories.
problem How do abstract categories like phonemes emerge from speech exposure?
method Used modeling techniques to test Memory-Based Learning and Error-Correction Learning.
result Error-Correction Learning models can learn abstractions, identifying phone inventory and grouping.
This paper explores what causal structures can be distinguished by observational and interventional probing schemes.
problem Identifying causal structures with latent variables using observational and interventional data.
method Investigates the power of different probing schemes (observation vs. intervention) to distinguish causal structures.
result Two causal structures are indistinguishable if they share the same mDAG structure.
In this work, we focus on fine-tuning an OpenAI GPT-2 pre-trained model for generating patent claims. GPT-2 has demonstrated impressive efficacy of pre-trained language models on various tasks, particularly coherent text generation. Patent claim language itself has rarely been explored in the past and poses a unique ch…
Paper analyzes why deeper layers of ViTs perform worse on out-of-distribution tasks.
problem Performance degradation of intermediate layers in ViTs under distribution shift.
method Extensive linear probing experiments across various benchmarks and fine-grained analysis of transformer modules.
result Probing feedforward network activations yields best performance under significant distribution shift.
Plex improves model reliability across vision and language tasks.
problem Improving model reliability in diverse decision-making tasks involving uncertainty and adaptation.
method Developed ViT-Plex and T5-Plex pretrained model extensions to evaluate and improve reliability across 40 datasets.
result Plex greatly improves state-of-the-art across reliability tasks, simplifying evaluation and performance.
A new approach to protein language models combines latent space prediction with masked language modeling.
problem Improving protein language models by predicting amino acid identities at masked positions.
method A variant of masked language modeling that predicts latent targets only at masked positions, retaining the MLM cross-entropy.
result The new approach outperforms pure masked language modeling on 11 out of 16 downstream tasks.
System guides freehand obstetric ultrasound probe movements.
problem Reducing operator expertise for standard plane acquisition.
method Artificial neural network trained on real scans.
result 88.8% accuracy for goal prediction, 90.9% for action prediction.
The ability of modeling the other agents, such as understanding their intentions and skills, is essential to an agent's interactions with other agents. Conventional agent modeling relies on passive observation from demonstrations. In this work, we propose an interactive agent modeling scheme enabled by encouraging an a…
CwA optimizes search performance by jointly learning a balanced database partition and a neural probing function.
problem Suboptimal search performance due to mismatched database and query distributions.
method CwA jointly learns a balanced database partition and a neural probing function to optimize search performance directly for the query distribution.
result CwA achieves up to 4.7x throughput over state-of-the-art methods at equal recall.
This work introduces a method to decompose uncertainty in in-context learning for large language models.
problem Understanding the sources of uncertainty in in-context learning for large language models.
method Variational uncertainty decomposition framework without sampling from latent parameter posterior.
result Quantitative and qualitative validation of decomposed epistemic and aleatoric uncertainties.
Improved scalability and interpretability in training data attribution.
problem Identifying which training data drives specific behaviors, especially unintended ones.
method Leveraging interpretable structures within the model to attribute model behavior to semantic directions, not individual test examples.
result Simple probe-based attribution methods are first-order approximations of Concept Influence that achieve comparable performance while being over an order-of-magnitude faster.
Galactica learns from scientific literature to help researchers.
problem Information overload in scientific literature makes it hard to find useful insights.
method Trained on a large corpus of scientific papers, reference material, and knowledge bases.
result Outperforms existing models on various scientific tasks, including LaTeX equations and mathematical reasoning.
Telescope detects LLM generated text by measuring token repetition probability.
problem Distinguishing LLM generated text from human writing.
method Telescope Perplexity, evaluating token repetition probability.
result Telescope Perplexity enables effective zero-shot LLM detection.
This paper focuses on the problem of estimating historical traffic volumes between sparsely-located traffic sensors, which transportation agencies need to accurately compute statewide performance measures. To this end, the paper examines applications of vehicle probe data, automatic traffic recorder counts, and neural …
Analyzes the concept of fields in classical and quantum physics.
problem Challenges in defining fields in classical and quantum physics.
method Uses groupoid description of quantum mechanics and categorical language.
result Fields as functors among groupoids of test particles and intrinsic system nature.
We study the statistical behavior of reasoning probes in a stylized model of iterative computation inspired by neural algorithmic reasoning. The underlying computation is given by a looped Boolean circuit whose graph is a perfect ν ν ν -ary tree ( ν ≥ 2 ν\ge 2 ν ≥ 2 ), with outputs recursively fed back as inputs across computation ro…
Language models fail to execute simple steps, showing gating and binding errors.
problem Procedural hallucinations in language models, failing to execute simple steps.
method Analyzed long-context binding tasks, identifying gating and binding errors.
result Procedural errors are due to gating and binding failures, with recency bias contributing to the latter.
HalluWorld benchmarks hallucinations in language models across diverse tasks.
problem Inconsistent benchmarks of hallucination in language models.
method Explicit reference-world formulation, synthetic and semi-synthetic environments, controlled variation of world complexity.
result Models struggle with perceptual hallucination but not multi-step state tracking.
Distribution grids currently lack comprehensive real-time metering. Nevertheless, grid operators require precise knowledge of loads and renewable generation to accomplish any feeder optimization task. At the same time, new grid technologies, such as solar photovoltaics and energy storage units are interfaced via invert…
Self-attention prefers sparse functions of input sequences, reducing sample complexity.
problem Understanding the inductive biases of self-attention in modeling long-range dependencies.
method Theoretical analysis and synthetic experiments to probe sample complexity of learning sparse functions with Transformers.
result Bounded-norm Transformer networks can represent sparse functions of the input sequence with logarithmic sample complexity.
This research discovers model architecture and training dataset characteristics through strategic input probing.
problem Discovering model architecture and training dataset characteristics in black box models.
method Structured input probes and model outputs are used to train a deep classifier for image and text classification.
result The approach successfully distinguishes between different image and text datasets and architectures.
There has been growing interest in using neural networks and deep learning techniques to create dialogue systems. Conversational recommendation is an interesting setting for the scientific exploration of dialogue with natural language as the associated discourse involves goal-driven dialogue that often transforms natur…
This work analyzes the role of data augmentation in self-supervised learning using RKHS approximation and regression.
problem Limited theoretical understanding of the role of data augmentation in self-supervised learning.
method Geometric characterization of the target function given by augmentation, proving generalization bounds.
result Two generalization bounds are derived, one free of model complexity, the other specific to near-optimal encoders.
Researchers create holographic super-embeddings for M5 and M2 branes.
problem No concrete examples of super-embeddings for M5 and M2 branes existed.
method Constructed explicit holographic super-embeddings of probe M5 and M2 branes into their super-AdS backgrounds.
result Explicit holographic super-embeddings of M5 and M2 branes were successfully constructed.