CondMTL improves toxicity detection by learning group-specific representations.
problem Algorithmic bias in toxic language detection across demographic groups.
method Conditional Multi-Task Learning (CondMTL) for demographic-specific tasks.
result CondMTL improves predictive recall for minority demographic groups.
The spectacular expansion of the Internet has led to the development of a new research problem in the field of natural language processing: automatic toxic comment detection, since many countries prohibit hate speech in public media. There is no clear and formal definition of hate, offensive, toxic and abusive speeches…
Paper develops a fast Bayesian method to predict toxic trades in financial transactions.
problem Predicting and managing toxic trades in financial transactions.
method PULSE: an online learning Bayesian procedure for neural networks.
result Neural networks trained with PULSE outperform traditional methods in predicting toxic trades.
To identify and classify toxic online commentary, the modern tools of data science transform raw text into key features from which either thresholding or learning algorithms can make predictions for monitoring offensive conversations. We systematically evaluate 62 classifiers representing 19 major algorithmic families …
UBM transfers bias mitigation from upstream to downstream tasks efficiently.
problem Bias in fine-tuned language models across various tasks.
method Apply bias mitigation to an upstream model, then fine-tune a downstream model on this mitigated model.
result UBM effects transfer to new downstream tasks, creating less biased models.
New framework controls statistical dispersion for high-stakes applications.
problem Understanding and controlling the dispersion of loss distributions in high-stakes applications.
method Simple yet flexible framework for distribution-free control of statistical dispersion measures.
result Proposed methods control statistical dispersion measures with societal implications.
Researchers develop method to protect against 'weight poisoning' attacks on pre-trained models.
problem The security threat of downloading untrusted pre-trained weights that can be manipulated after fine-tuning.
method RIPPLe regularization method and Embedding Surgery initialization procedure.
result Demonstrated that weight poisoning attacks are possible even with limited knowledge of the dataset and fine-tuning procedure.
Bayesian framework improves minority class performance in class-imbalanced data.
problem Class imbalance in predictive toxicology models.
method Weighted likelihood approach modifying likelihood function weights inversely proportional to class proportions.
result Improves balanced accuracy and sensitivity for minority class (toxic compounds).
We propose the development of a prediction market for forecasting prices for "toxic assets" to be transferred from Irish banks to the National Asset Management Agency (NAMA). Such a market allows market participants to assume a stake in a security whose value is tied to a future event. We propose that securities are cr…
Timely assessment of compound toxicity is one of the biggest challenges facing the pharmaceutical industry today. A significant proportion of compounds identified as potential leads are ultimately discarded due to the toxicity they induce. In this paper, we propose a novel machine learning approach for the prediction o…
Everyday we are exposed to various chemicals via food additives, cleaning and cosmetic products and medicines -- and some of them might be toxic. However testing the toxicity of all existing compounds by biological experiments is neither financially nor logistically feasible. Therefore the government agencies NIH, EPA …
We refine toxicity bounds for dynamic liquidation incentives in CP-AMM systems.
problem Ensuring stability in dynamic liquidation incentives in automated market makers.
method Derived state-dependent toxicity bounds for dynamic liquidation incentives, reconciling them with CP-AMM price dynamics.
result State-dependent bounds and liquidity-depth-only condition for dynamic liquidation incentives.
Prediction of toxicity levels of chemical compounds is an important issue in Quantitative Structure-Activity Relationship (QSAR) modeling. Although toxicity prediction has achieved significant progress in recent times through deep learning, prediction accuracy levels obtained by even very recent methods are not yet ver…
Model optimizes trading strategy with unobservable toxicity.
problem Maximizing daily trading profit with unobservable toxicity.
method Formulated as a partially observable stochastic control problem, solved in two steps.
result P&L performance gap is negligible (0.01%) in all scenarios.
This research secures deployed sentiment analysis models by identifying and defending against attack vectors.
problem Securing deployed machine learning models, particularly sentiment analysis systems, from adversarial attacks.
method BAD (Build, Attack, Defend) Architecture, evaluating two implementations.
result Demonstrated a viable methodology for securing machine learning models in production.
Mitigates bias in text classification by weighting instances.
problem Unintended biases in text classification datasets based on demographic terms.
method Instance weighting to recover non-discrimination distribution.
result Effective mitigation of unintended biases without sacrificing generalization.
SEEDA optimizes dose allocation in clinical trials to balance efficacy and safety.
problem Complex relationships between efficacy and toxicity in new drug trials.
method Adaptive clinical trial methodology that maximizes cumulative efficacy while ensuring safety constraints.
result SEEDA outperforms existing methods in finding optimal doses with higher success rates and fewer patients.
Toxicity prediction of chemical compounds is a grand challenge. Lately, it achieved significant progress in accuracy but using a huge set of features, implementing a complex blackbox technique such as a deep neural network, and exploiting enormous computational resources. In this paper, we strongly argue for the models…
New method designs antimicrobial peptides with high potency and low toxicity.
problem Designing potent antimicrobial drugs with low toxicity.
method CLaSS method using deep generative autoencoder and atomistic simulations.
result Design and synthesis of two novel AMPs with high potency and low toxicity.
Work shows hallucination detection by LLMs is impossible without expert feedback.
problem Detecting hallucinations in LLMs is theoretically impossible without expert-labeled feedback.
method Investigated hallucination detection using a theoretical framework inspired by language identification.
result Automated hallucination detection is impossible for most language collections without expert-labeled feedback.
We develop a statistical framework to benchmark and select large language models based on their risks.
problem Benchmarking and selecting large language models based on their associated risks.
method A distributional framework using first and second order stochastic dominance, linked to mean-risk models in finance.
result Formalizes a risk-aware approach for model selection, balancing risk and utility.
Framework ensures alignment between humans and machines in LLMs.
problem Human-machine misalignment in LLMs scoring mechanisms.
method Lightweight calibration framework for blackbox models.
result Provably guarantees alignment between humans and machines.
LLM Pro Finance Suite enhances financial NLP with instruction-tuned models.
problem Limited NLP capabilities for financial tasks in generalist models.
method Instruction-tuned large language models fine-tuned on financial data.
result Consistent improvement over state-of-the-art baselines in finance tasks.
Combining watermark and non-watermark detectors improves LLM detection.
problem Challenges in detecting watermarked language models due to limited entropy.
method Investigates hybrid schemes combining watermark and non-watermark detectors.
result Hybrid schemes outperform either class of detector under various conditions.
System identifies language of transliterated text.
problem Users struggle to understand non-native language transliterated text.
method Feature extraction of phonetic syllables using LSTM network.
result System accurately identifies language of transliterated text.
Proposes a method to generate text that adheres to logical constraints.
problem Generating text that respects logical constraints is hard for autoregressive models.
method Bayesian conditioning to draw samples subject to a constraint, considering the entire sequence and inducing a local, factorized distribution.
result Our approach generates samples that closely approximate the target distribution and are guaranteed to satisfy the constraints.
RoBERTa model detects counterfactual statements in text.
problem Detecting and extracting counterfactual statements from text.
method Used RoBERTa language representation model for both subtasks.
result RoBERTa achieved top performance in both subtasks at SemEval-2020.
Multi-language speech datasets are scarce and often have small sample sizes in the medical domain. Robust transfer of linguistic features across languages could improve rates of early diagnosis and therapy for speakers of low-resource languages when detecting health conditions from speech. We utilize out-of-domain, unp…
New approach extracts AI model representations for steering and monitoring.
problem Understanding and steering AI models' internal knowledge representation.
method Feature learning advances for scalable linear concept representations.
result Concept representations enable steering and monitoring of AI models.
Many modern data sets are sampled with error from complex high-dimensional surfaces. Methods such as tensor product splines or Gaussian processes are effective/well suited for characterizing a surface in two or three dimensions but may suffer from difficulties when representing higher dimensional surfaces. Motivated by…
Study detects SLI in children from spontaneous narrative transcripts.
problem Detecting Specific Language Impairment (SLI) in children.
method Three-stage pipeline: feature extraction, dimensionality reduction, and classification.
result 97.13% accuracy in identifying SLI from transcripts.
Toxicity analysis and prediction are of paramount importance to human health and environmental protection. Existing computational methods are built from a wide variety of descriptors and regressors, which makes their performance analysis difficult. For example, deep neural network (DNN), a successful approach in many o…
A new watermarking method corrects bias in language models using maximal coupling.
problem Correcting bias in language model token distributions.
method Maximal coupling to balance bias correction and text quality.
result Outperforms prior techniques in preserving text quality and detectability.
We provide a comprehensive investigation of different custom and off-the-shelf architectures as well as different approaches to generating feature vectors for offensive language detection. We also show that these approaches work well on small and noisy datasets such as on the Offensive Language Identification Dataset (…
Automatically jailbreaks LLMs with black-box access.
problem Generating harmful content from black-box LLMs.
method Automated method using an attacker LLM to refine prompts.
result Generates jailbreaks for over 80% of prompts.
Model detects depression from transcribed interviews using affective language.
problem Detecting depression from transcribed clinical interviews.
method Hierarchical Attention Network with affective conditioning.
result Model achieves state-of-the-art F1 scores in depression detection.
This study uses NLP to detect financial risks from documents.
problem Detecting and predicting financial risks in documents.
method NLP model design, text preprocessing, feature extraction, machine learning.
result NLP model effectively identifies and predicts financial risks.
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.
Language is dynamic, constantly evolving and adapting with respect to time, domain or topic. The adaptability of language is an active research area, where researchers discover social, cultural and domain-specific changes in language using distributional tools such as word embeddings. In this paper, we introduce the gl…
Humor is a unique and creative communicative behavior displayed during social interactions. It is produced in a multimodal manner, through the usage of words (text), gestures (vision) and prosodic cues (acoustic). Understanding humor from these three modalities falls within boundaries of multimodal language; a recent r…
Method detects if text is generated by a language model with watermarks.
problem Detecting if text is generated by a language model with watermarks.
method Randomization tests and change point detection techniques.
result Method ensures Type I and Type II error control and accurately identifies watermarked sub-strings.
Decoding strategies often exclude human-like tokens, creating a detectable gap in generated text.
problem Decoding strategies exclude contextually appropriate but statistically rare tokens, creating a detectable gap in generated text.
method Analysis of 1.8 million texts across 8 language models, 5 decoding strategies, and 53 hyperparameter configurations.
result 8-18% of human-selected tokens fall outside typical truncation boundaries, indicating a detectable gap.
TLMG4Eth combines language and graph models for Ethereum fraud detection.
problem Current fraud detection methods fail to consider semantic and similarity patterns in Ethereum transactions.
method TLMG4Eth uses a transaction language model and graph-based methods to capture semantic, similarity, and structural features.
result TLMG4Eth detects anomalies in Ethereum transactions more effectively than existing methods.
This paper improves bond market making by adjusting hit-ratios for client flow quality.
problem Economic misleading of raw hit-ratios in corporate bond market making.
method Stochastic-control framework with residual-quality-adjusted hit-ratio.
result Optimal quotes decompose into various components, improving service/economics frontier.
LLMs struggle with zero-shot annotation tasks due to model-internalized priors.
problem Impact of model-internalized priors on LLM performance in zero-shot annotation tasks.
method Investigated three dimensions: familiarity, decision stickiness, and susceptibility to misaligned task definitions.
result Nearly two-thirds of zero-shot errors are resistant to correction, with a rescue rate of 34.8%. Definition-Specific Familiarity (DSF) shows a positive association with model performance.
Visual spoofing bypasses spam filters and plagiarism detection.
problem Vulnerability in spam filters that can be exploited by visually similar but differently encoded characters.
method Replaces characters with visually similar but differently encoded characters from a different alphabet.
result Spammers can create messages that bypass existing spam filters.
The study detects deceptive language in business communication using AI.
problem Deceptive language in business communication.
method Combining classical rhetoric, communication psychology, and linguistic theory with computational textual analysis and transformer models.
result Detection accuracies of over 99% achieved in controlled settings.
CogMol designs novel drug-like molecules for SARS-CoV-2 targets.
problem Designing efficient drugs for novel viral proteins.
method End-to-end framework combining VAE, controlled sampling, and predictors.
result Highly selective and affinity molecules for SARS-CoV-2 targets.