arXiv subjects

Zohar Karnin

Publications and source records attributed to Zohar Karnin.

At least 19 recordsLinked to original sources

arXiv2026

Optimal Budgeted Adaptation of Large Language Models

The trade-off between labeled data availability and downstream accuracy remains a central challenge in fine-tuning large language models (LLMs). We propose a principled framework for \emph{budget-aware supervised fine-tuning} by casting LLM adaptation as a contextual Stackelberg game. In our formulation, the learner (leader) commits to a scoring policy and a label-querying strategy, while an adaptive environment (follower) selects challenging supervised alternatives in response. To explicitly address label efficiency, we incorporate a finite supervision budget directly into the learning objective. Our algorithm operates in the full-feedback regime and achieves $\tilde{O}(d\sqrt{T})$ regret under standard linear contextual assumptions. We extend the framework with a Largest-Latency-First (LLF) confidence gate that selectively queries labels, achieving a budget-aware regret bound of $\tilde{O}(\sqrt{dB} + c\sqrt{B})$ with $B=βT$.

cs.LG↗

arXiv2026

Linguistic and Argument Diversity in Synthetic Data for Function-Calling Agents

The construction of function calling agents has emerged as a promising avenue for extending model capabilities. A major challenge for this task is obtaining high quality diverse data for training. Prior work emphasizes diversity in functions, invocation patterns, and interaction turns, yet linguistic diversity of requests and coverage of arguments (e.g., \texttt{city\_name}, \texttt{stock\_ticker}) remain underexplored. We propose a method that generates synthetic datasets via optimizing general-purpose diversity metrics across both queries and arguments, without relying on hand-crafted rules or taxonomies, making it robust to different usecases. We demonstrate the effectiveness of our technique via both intrinsic and extrinsic testing, comparing it to SoTA data generation methods. We show a superiority over baselines in terms of diversity, while keeping comparable correctness. Additionally, when used as a training set, the model resulting from our dataset exhibits superior performance compared to analogous models based on the baseline data generation methods in out-of-distribution performance. In particular, we achieve an $7.4\%$ increase in accuracy on the BFCL benchmark compared to similar counterparts.

cs.CL↗

arXiv2025

The Cross-Lingual Cost: Retrieval Biases in RAG over Arabic-English Corpora

Cross-lingual retrieval-augmented generation (RAG) is a critical capability for retrieving and generating answers across languages. Prior work in this context has mostly focused on generation and relied on benchmarks derived from open-domain sources, most notably Wikipedia. In such settings, retrieval challenges often remain hidden due to language imbalances, overlap with pretraining data, and memorized content. To address this gap, we study Arabic-English RAG in a domain-specific setting using benchmarks derived from real-world corporate datasets. Our benchmarks include all combinations of languages for the user query and the supporting document, drawn independently and uniformly at random. This enables a systematic study of multilingual retrieval behavior. Our findings reveal that retrieval is a critical bottleneck in cross-lingual domain-specific scenarios, with substantial performance drops occurring when the user query and supporting document languages differ. A key insight is that these failures stem primarily from the retriever's difficulty in ranking documents across languages. Finally, we propose two simple retrieval strategies that address this source of failure by enforcing equal retrieval from both languages or by translating the query, resulting in substantial improvements in cross-lingual and overall performance. These results highlight meaningful opportunities for improving multilingual retrieval, particularly in practical, real-world RAG applications.

cs.CL↗

arXiv2025

The Distracting Effect: Understanding Irrelevant Passages in RAG

A well-known issue with Retrieval Augmented Generation (RAG) is that retrieved passages that are irrelevant to the query sometimes distract the answer-generating LLM, causing it to provide an incorrect response. In this paper, we shed light on this core issue and formulate the distracting effect of a passage w.r.t. a query (and an LLM). We provide a quantifiable measure of the distracting effect of a passage and demonstrate its robustness across LLMs. Our research introduces novel methods for identifying and using hard distracting passages to improve RAG systems. By fine-tuning LLMs with these carefully selected distracting passages, we achieve up to a 7.5% increase in answering accuracy compared to counterparts fine-tuned on conventional RAG datasets. Our contribution is two-fold: first, we move beyond the simple binary classification of irrelevant passages as either completely unrelated vs. distracting, and second, we develop and analyze multiple methods for finding hard distracting passages. To our knowledge, no other research has provided such a comprehensive framework for identifying and utilizing hard distracting passages.

cs.CL↗

arXiv2025

Generating Diverse Q&A Benchmarks for RAG Evaluation with DataMorgana

Evaluating Retrieval-Augmented Generation (RAG) systems, especially in domain-specific contexts, requires benchmarks that address the distinctive requirements of the applicative scenario. Since real data can be hard to obtain, a common strategy is to use LLM-based methods to generate synthetic data. Existing solutions are general purpose: given a document, they generate a question to build a Q&A pair. However, although the generated questions can be individually good, they are typically not diverse enough to reasonably cover the different ways real end-users can interact with the RAG system. We introduce here DataMorgana, a tool for generating highly customizable and diverse synthetic Q&A benchmarks tailored to RAG applications. DataMorgana enables detailed configurations of user and question categories and provides control over their distribution within the benchmark. It uses a lightweight two-stage process, ensuring efficiency and fast iterations, while generating benchmarks that reflect the expected traffic. We conduct a thorough line of experiments, showing quantitatively and qualitatively that DataMorgana surpasses existing tools and approaches in producing lexically, syntactically, and semantically diverse question sets across domain-specific and general-knowledge corpora. DataMorgana will be made available to selected teams in the research community, as first beta testers, in the context of the upcoming SIGIR'2025 LiveRAG challenge to be announced in early February 2025.

cs.CL↗

arXiv2024

Quality Matters: Evaluating Synthetic Data for Tool-Using LLMs

Training large language models (LLMs) for external tool usage is a rapidly expanding field, with recent research focusing on generating synthetic data to address the shortage of available data. However, the absence of systematic data quality checks poses complications for properly training and testing models. To that end, we propose two approaches for assessing the reliability of data for training LLMs to use external tools. The first approach uses intuitive, human-defined correctness criteria. The second approach uses a model-driven assessment with in-context evaluation. We conduct a thorough evaluation of data quality on two popular benchmarks, followed by an extrinsic evaluation that showcases the impact of data quality on model performance. Our results demonstrate that models trained on high-quality data outperform those trained on unvalidated data, even when trained with a smaller quantity of data. These findings empirically support the significance of assessing and ensuring the reliability of training data for tool-using LLMs.

cs.LG↗

arXiv2023

Representation Projection Invariance Mitigates Representation Collapse

Fine-tuning contextualized representations learned by pre-trained language models remains a prevalent practice in NLP. However, fine-tuning can lead to representation degradation (also known as representation collapse), which may result in instability, sub-optimal performance, and weak generalization. In this paper, we propose Representation Projection Invariance (REPINA), a novel regularization method to maintain the information content of representation and reduce representation collapse during fine-tuning by discouraging undesirable changes in the representations. We study the empirical behavior of the proposed regularization in comparison to 5 comparable baselines across 13 language understanding tasks (GLUE benchmark and six additional datasets). When evaluating in-domain performance, REPINA consistently outperforms other baselines on most tasks (10 out of 13). We also demonstrate its effectiveness in few-shot settings and robustness to label perturbation. As a by-product, we extend previous studies of representation collapse and propose several metrics to quantify it. Our empirical findings show that our approach is significantly more effective at mitigating representation collapse.

cs.CL↗

arXiv2023

Uncovering Drift in Textual Data: An Unsupervised Method for Detecting and Mitigating Drift in Machine Learning Models

Drift in machine learning refers to the phenomenon where the statistical properties of data or context, in which the model operates, change over time leading to a decrease in its performance. Therefore, maintaining a constant monitoring process for machine learning model performance is crucial in order to proactively prevent any potential performance regression. However, supervised drift detection methods require human annotation and consequently lead to a longer time to detect and mitigate the drift. In our proposed unsupervised drift detection method, we follow a two step process. Our first step involves encoding a sample of production data as the target distribution, and the model training data as the reference distribution. In the second step, we employ a kernel-based statistical test that utilizes the maximum mean discrepancy (MMD) distance metric to compare the reference and target distributions and estimate any potential drift. Our method also identifies the subset of production data that is the root cause of the drift. The models retrained using these identified high drift samples show improved performance on online customer experience quality metrics.

cs.CL↗

arXiv2023

Relative Error Streaming Quantiles

Estimating ranks, quantiles, and distributions over streaming data is a central task in data analysis and monitoring. Given a stream of $n$ items from a data universe equipped with a total order, the task is to compute a sketch (data structure) of size polylogarithmic in $n$. Given the sketch and a query item $y$, one should be able to approximate its rank in the stream, i.e., the number of stream elements smaller than or equal to $y$. Most works to date focused on additive $\varepsilon n$ error approximation, culminating in the KLL sketch that achieved optimal asymptotic behavior. This paper investigates multiplicative $(1\pm\varepsilon)$-error approximations to the rank. Practical motivation for multiplicative error stems from demands to understand the tails of distributions, and hence for sketches to be more accurate near extreme values. The most space-efficient algorithms due to prior work store either $O(\log(\varepsilon^2 n)/\varepsilon^2)$ or $O(\log^3(\varepsilon n)/\varepsilon)$ universe items. We present a randomized sketch storing $O(\log^{1.5}(\varepsilon n)/\varepsilon)$ items that can $(1\pm\varepsilon)$-approximate the rank of each universe item with high constant probability; this space bound is within an $O(\sqrt{\log(\varepsilon n)})$ factor of optimal. Our algorithm does not require prior knowledge of the stream length and is fully mergeable, rendering it suitable for parallel and distributed computing environments.

cs.DS↗

arXiv2022

Amazon SageMaker Model Monitor: A System for Real-Time Insights into Deployed Machine Learning Models

With the increasing adoption of machine learning (ML) models and systems in high-stakes settings across different industries, guaranteeing a model's performance after deployment has become crucial. Monitoring models in production is a critical aspect of ensuring their continued performance and reliability. We present Amazon SageMaker Model Monitor, a fully managed service that continuously monitors the quality of machine learning models hosted on Amazon SageMaker. Our system automatically detects data, concept, bias, and feature attribution drift in models in real-time and provides alerts so that model owners can take corrective actions and thereby maintain high quality models. We describe the key requirements obtained from customers, system design and architecture, and methodology for detecting different types of drift. Further, we provide quantitative evaluations followed by use cases, insights, and lessons learned from more than two years of production deployment.

cs.LG↗

arXiv2022

Pyramid-BERT: Reducing Complexity via Successive Core-set based Token Selection

Transformer-based language models such as BERT have achieved the state-of-the-art performance on various NLP tasks, but are computationally prohibitive. A recent line of works use various heuristics to successively shorten sequence length while transforming tokens through encoders, in tasks such as classification and ranking that require a single token embedding for prediction. We present a novel solution to this problem, called Pyramid-BERT where we replace previously used heuristics with a {\em core-set} based token selection method justified by theoretical results. The core-set based token selection technique allows us to avoid expensive pre-training, gives a space-efficient fine tuning, and thus makes it suitable to handle longer sequence lengths. We provide extensive experiments establishing advantages of pyramid BERT over several baselines and existing works on the GLUE benchmarks and Long Range Arena datasets.

cs.CL↗

arXiv2021

Improving Early Sepsis Prediction with Multi Modal Learning

Sepsis is a life-threatening disease with high morbidity, mortality and healthcare costs. The early prediction and administration of antibiotics and intravenous fluids is considered crucial for the treatment of sepsis and can save potentially millions of lives and billions in health care costs. Professional clinical care practitioners have proposed clinical criterion which aid in early detection of sepsis; however, performance of these criterion is often limited. Clinical text provides essential information to estimate the severity of the sepsis in addition to structured clinical data. In this study, we explore how clinical text can complement structured data towards early sepsis prediction task. In this paper, we propose multi modal model which incorporates both structured data in the form of patient measurements as well as textual notes on the patient. We employ state-of-the-art NLP models such as BERT and a highly specialized NLP model in Amazon Comprehend Medical to represent the text. On the MIMIC-III dataset containing records of ICU admissions, we show that by using these notes, one achieves an improvement of 6.07 points in a standard utility score for Sepsis prediction and 2.89% in AUROC score. Our methods significantly outperforms a clinical criteria suggested by experts, qSOFA, as well as the winning model of the PhysioNet Computing in Cardiology Challenge for predicting Sepsis.

cs.CL↗

arXiv2021

GANMEX: One-vs-One Attributions Guided by GAN-based Counterfactual Explanation Baselines

Attribution methods have been shown as promising approaches for identifying key features that led to learned model predictions. While most existing attribution methods rely on a baseline input for performing feature perturbations, limited research has been conducted to address the baseline selection issues. Poor choices of baselines limit the ability of one-vs-one (1-vs-1) explanations for multi-class classifiers, which means the attribution methods were not able to explain why an input belongs to its original class but not the other specified target class. 1-vs-1 explanation is crucial when certain classes are more similar than others, e.g. two bird types among multiple animals, by focusing on key differentiating features rather than shared features across classes. In this paper, we present GAN-based Model EXplainability (GANMEX), a novel approach applying Generative Adversarial Networks (GAN) by incorporating the to-be-explained classifier as part of the adversarial networks. Our approach effectively selects the counterfactual baseline as the closest realistic sample belong to the target class, which allows attribution methods to provide true 1-vs-1 explanations. We showed that GANMEX baselines improved the saliency maps and led to stronger performance on perturbation-based evaluation metrics over the existing baselines. Existing attribution results are known for being insensitive to model randomization, and we demonstrated that GANMEX baselines led to better outcome under the cascading randomization of the model.

cs.LG↗

arXiv2020

Amazon SageMaker Autopilot: a white box AutoML solution at scale

AutoML systems provide a black-box solution to machine learning problems by selecting the right way of processing features, choosing an algorithm and tuning the hyperparameters of the entire pipeline. Although these systems perform well on many datasets, there is still a non-negligible number of datasets for which the one-shot solution produced by each particular system would provide sub-par performance. In this paper, we present Amazon SageMaker Autopilot: a fully managed system providing an automated ML solution that can be modified when needed. Given a tabular dataset and the target column name, Autopilot identifies the problem type, analyzes the data and produces a diverse set of complete ML pipelines including feature preprocessing and ML algorithms, which are tuned to generate a leaderboard of candidate models. In the scenario where the performance is not satisfactory, a data scientist is able to view and edit the proposed ML pipelines in order to infuse their expertise and business knowledge without having to revert to a fully manual solution. This paper describes the different components of Autopilot, emphasizing the infrastructure choices that allow scalability, high quality models, editable ML pipelines, consumption of artifacts of offline meta-learning, and a convenient integration with the entire SageMaker suite allowing these trained models to be used in a production setting.

cs.LG↗

arXiv2020

TabTransformer: Tabular Data Modeling Using Contextual Embeddings

We propose TabTransformer, a novel deep tabular data modeling architecture for supervised and semi-supervised learning. The TabTransformer is built upon self-attention based Transformers. The Transformer layers transform the embeddings of categorical features into robust contextual embeddings to achieve higher prediction accuracy. Through extensive experiments on fifteen publicly available datasets, we show that the TabTransformer outperforms the state-of-the-art deep learning methods for tabular data by at least 1.0% on mean AUC, and matches the performance of tree-based ensemble models. Furthermore, we demonstrate that the contextual embeddings learned from TabTransformer are highly robust against both missing and noisy data features, and provide better interpretability. Lastly, for the semi-supervised setting we develop an unsupervised pre-training procedure to learn data-driven contextual embeddings, resulting in an average 2.1% AUC lift over the state-of-the-art methods.

cs.LG↗

arXiv2020

Practical and sample efficient zero-shot HPO

Zero-shot hyperparameter optimization (HPO) is a simple yet effective use of transfer learning for constructing a small list of hyperparameter (HP) configurations that complement each other. That is to say, for any given dataset, at least one of them is expected to perform well. Current techniques for obtaining this list are computationally expensive as they rely on running training jobs on a diverse collection of datasets and a large collection of randomly drawn HPs. This cost is especially problematic in environments where the space of HPs is regularly changing due to new algorithm versions, or changing architectures of deep networks. We provide an overview of available approaches and introduce two novel techniques to handle the problem. The first is based on a surrogate model and adaptively chooses pairs of dataset, configuration to query. The second, for settings where finding, tuning and testing a surrogate model is problematic, is a multi-fidelity technique combining HyperBand with submodular optimization. We benchmark our methods experimentally on five tasks (XGBoost, LightGBM, CatBoost, MLP and AutoML) and show significant improvement in accuracy compared to standard zero-shot HPO with the same training budget. In addition to contributing new algorithms, we provide an extensive study of the zero-shot HPO technique resulting in (1) default hyper-parameters for popular algorithms that would benefit the community using them, (2) massive lookup tables to further the research of hyper-parameter tuning.

stat.ML↗

arXiv2020

An Empirical Process Approach to the Union Bound: Practical Algorithms for Combinatorial and Linear Bandits

This paper proposes near-optimal algorithms for the pure-exploration linear bandit problem in the fixed confidence and fixed budget settings. Leveraging ideas from the theory of suprema of empirical processes, we provide an algorithm whose sample complexity scales with the geometry of the instance and avoids an explicit union bound over the number of arms. Unlike previous approaches which sample based on minimizing a worst-case variance (e.g. G-optimal design), we define an experimental design objective based on the Gaussian-width of the underlying arm set. We provide a novel lower bound in terms of this objective that highlights its fundamental role in the sample complexity. The sample complexity of our fixed confidence algorithm matches this lower bound, and in addition is computationally efficient for combinatorial classes, e.g. shortest-path, matchings and matroids, where the arm sets can be exponentially large in the dimension. Finally, we propose the first algorithm for linear bandits in the the fixed budget setting. Its guarantee matches our lower bound up to logarithmic factors.

cs.LG↗

arXiv2020

PruneNet: Channel Pruning via Global Importance

Channel pruning is one of the predominant approaches for accelerating deep neural networks. Most existing pruning methods either train from scratch with a sparsity inducing term such as group lasso, or prune redundant channels in a pretrained network and then fine tune the network. Both strategies suffer from some limitations: the use of group lasso is computationally expensive, difficult to converge and often suffers from worse behavior due to the regularization bias. The methods that start with a pretrained network either prune channels uniformly across the layers or prune channels based on the basic statistics of the network parameters. These approaches either ignore the fact that some CNN layers are more redundant than others or fail to adequately identify the level of redundancy in different layers. In this work, we investigate a simple-yet-effective method for pruning channels based on a computationally light-weight yet effective data driven optimization step that discovers the necessary width per layer. Experiments conducted on ILSVRC-$12$ confirm effectiveness of our approach. With non-uniform pruning across the layers on ResNet-$50$, we are able to match the FLOP reduction of state-of-the-art channel pruning results while achieving a $0.98\%$ higher accuracy. Further, we show that our pruned ResNet-$50$ network outperforms ResNet-$34$ and ResNet-$18$ networks, and that our pruned ResNet-$101$ outperforms ResNet-$50$.

cs.LG↗