Tanya Roosta

Peyman Passban, Tanya Roosta, Rahul Gupta et al. · amazon-science

Training mixed-domain translation models is a complex task that demands tailored architectures and costly data preparation techniques. In this work, we leverage federated learning (FL) in order to tackle the problem. Our investigation demonstrates that with slight modifications in the training process, neural machine translation (NMT) engines can be easily adapted when an FL-based aggregation is applied to fuse different domains. Experimental results also show that engines built via FL are able to perform on par with state-of-the-art baselines that rely on centralized training techniques. We evaluate our hypothesis in the presence of five datasets with different sizes, from different domains, to translate from German into English and discuss how FL and NMT can mutually benefit from each other. In addition to providing benchmarking results on the union of FL and NMT, we also propose a novel technique to dynamically control the communication bandwidth by selecting impactful parameters during FL updates. This is a significant achievement considering the large size of NMT engines that need to be exchanged between FL parties.

Aishwarya Sarkar, Sayan Ghosh, Nathan Tallent et al.

Large-scale Graph Neural Networks (GNNs) are typically trained by sampling a vertex's neighbors to a fixed distance. Because large input graphs are distributed, training requires frequent irregular communication that stalls forward progress. Moreover, fetched data changes with graph, graph distribution, sample and batch parameters, and caching polices. Consequently, any static prefetching method will miss crucial opportunities to adapt to different dynamic conditions. In this paper, we introduce Rudder, a software module embedded in the state-of-the-art AWS DistDGL framework, to autonomously prefetch remote nodes and minimize communication. Rudder's adaptation contrasts with both standard heuristics and traditional ML classifiers. We observe that the generative AI found in contemporary Large Language Models (LLMs) exhibits emergent properties like In-Context Learning (ICL) for zero-shot tasks, with logical multi-step reasoning. We find this behavior well-suited for adaptive control even with substantial undertraining. Evaluations using standard datasets and unseen configurations on the NERSC Perlmutter supercomputer show up to 91% improvement in end-to-end training performance over baseline DistDGL (no prefetching), and an 82% improvement over static prefetching, reducing communication by over 50%. Our code is available at https://github.com/aishwaryyasarkar/rudder-llm-agent.

Manas Mohanty, Tanya Roosta, Peyman Passban · amazon-science

Deep neural networks (DNNs) have improved NLP tasks significantly, but training and maintaining such networks could be costly. Model compression techniques, such as, knowledge distillation (KD), have been proposed to address the issue; however, the compression process could be lossy. Motivated by this, our work investigates how a distilled student model differs from its teacher, if the distillation process causes any information losses, and if the loss follows a specific pattern. Our experiments aim to shed light on the type of tasks might be less or more sensitive to KD by reporting data points on the contribution of different factors, such as the number of layers or attention heads. Results such as ours could be utilized when determining effective and efficient configurations to achieve optimal information transfers between larger (teacher) and smaller (student) models.

Md Anwar Hossen, Fatema Siddika, Juan Pablo Munoz et al.

Large language models (LLMs) can acquire new capabilities through fine-tuning, but continual adaptation often leads to catastrophic forgetting. We propose CRAFT, a continual learning framework that avoids updating model weights by instead learning low-rank interventions on hidden representations. CRAFT proceeds in three stages: it first routes each task to a group of similar tasks based on output-distribution divergence; it then fine-tunes the model using a Kullback-Leibler (KL) divergence against the group's prior state, which directly controls forgetting and determines convergence; finally, it merges interventions for the updated task into the shared representation using the same KL signal. This design unifies routing, regularization, and merging through a single KL-based objective. CRAFT improves overall performance and reduces forgetting compared to strong LoRA-based approaches across multiple benchmarks and model scales, while remaining robust to task ordering. These results suggest that controlling adaptation in representation space, guided by output-space divergence, provides a scalable and principled approach to continual learning in LLMs.

Yu Lu Liu, Hyokun Yun, Tanya Roosta et al.

There is growing interest in exploring user simulation as an alternative to gathering and scoring real user-chatbot interactions for AI chatbot evaluation. For this purpose, it is important to ensure the realism of the simulation, i.e., the extent to which simulated dialogues reflect real dialogues users have with chatbots. Most existing methods evaluating simulation realism produce coarse quality signal and remain solely at the level of individual dialogues. To support more rigorous evaluation in this area, we propose realsim, an evaluation framework that enables practitioners to take a distributional view of real vs. simulated dialogues along 8 dimensions, covering attributes related to the communicative functions of the interaction, user states, and the surface form of user messages. We then instantiate the framework with a curated dataset of 1K multi-turn task-focused real user-chatbot dialogues that cover 16 domains of chatbot applications. Overall, we find that simulated users tend to struggle at capturing communication frictions that real users introduce to interactions, which could make evaluations based on such simulations overly optimistic. We also observe variability in performance across different domains, which may indicate a need for domain-specific user simulators.

Preetam Prabhu Srikar Dammu, Himanshu Naidu, Mouly Dewan et al.

In the midst of widespread misinformation and disinformation through social media and the proliferation of AI-generated texts, it has become increasingly difficult for people to validate and trust information they encounter. Many fact-checking approaches and tools have been developed, but they often lack appropriate explainability or granularity to be useful in various contexts. A text validation method that is easy to use, accessible, and can perform fine-grained evidence attribution has become crucial. More importantly, building user trust in such a method requires presenting the rationale behind each prediction, as research shows this significantly influences people's belief in automated systems. Localizing and bringing users' attention to the specific problematic content is also paramount, instead of providing simple blanket labels. In this paper, we present ClaimVer, a human-centric framework tailored to meet users' informational and verification needs by generating rich annotations and thereby reducing cognitive load. Designed to deliver comprehensive evaluations of texts, it highlights each claim, verifies it against a trusted knowledge graph (KG), presents the evidence, and provides succinct, clear explanations for each claim prediction. Finally, our framework introduces an attribution score, enhancing applicability across a wide range of downstream tasks.

Maryam Amirizaniani, Jihan Yao, Adrian Lavergne et al.

As Large Language Models (LLMs) become more pervasive across various users and scenarios, identifying potential issues when using these models becomes essential. Examples of such issues include: bias, inconsistencies, and hallucination. Although auditing the LLM for these problems is often warranted, such a process is neither easy nor accessible for most. An effective method is to probe the LLM using different versions of the same question. This could expose inconsistencies in its knowledge or operation, indicating potential for bias or hallucination. However, to operationalize this auditing method at scale, we need an approach to create those probes reliably and automatically. In this paper we propose the LLMAuditor framework which is an automatic, and scalable solution, where one uses a different LLM along with human-in-the-loop (HIL). This approach offers verifiability and transparency, while avoiding circular reliance on the same LLM, and increasing scientific rigor and generalizability. Specifically, LLMAuditor includes two phases of verification using humans: standardized evaluation criteria to verify responses, and a structured prompt template to generate desired probes. A case study using questions from the TruthfulQA dataset demonstrates that we can generate a reliable set of probes from one LLM that can be used to audit inconsistencies in a different LLM. This process is enhanced by our structured prompt template with HIL, which not only boosts the reliability of our approach in auditing but also yields the delivery of less hallucinated results. The novelty of our research stems from the development of a comprehensive, general-purpose framework that includes a HIL verified prompt template for auditing responses generated by LLMs.

Suyu Ye, Haojun Shi, Darren Shih et al.

To achieve successful assistance with long-horizon web-based tasks, AI agents must be able to sequentially follow real-world user instructions over a long period. Unlike existing web-based agent benchmarks, sequential instruction following in the real world poses significant challenges beyond performing a single, clearly defined task. For instance, real-world human instructions can be ambiguous, require different levels of AI assistance, and may evolve over time, reflecting changes in the user's mental state. To address this gap, we introduce RealWebAssist, a novel benchmark designed to evaluate sequential instruction-following in realistic scenarios involving long-horizon interactions with the web, visual GUI grounding, and understanding ambiguous real-world user instructions. RealWebAssist includes a dataset of sequential instructions collected from real-world human users. Each user instructs a web-based assistant to perform a series of tasks on multiple websites. A successful agent must reason about the true intent behind each instruction, keep track of the mental state of the user, understand user-specific routines, and ground the intended tasks to actions on the correct GUI elements. Our experimental results show that state-of-the-art models struggle to understand and ground user instructions, posing critical challenges in following real-world user instructions for long-horizon web assistance.

Maryam Amirizaniani, Elias Martin, Tanya Roosta et al.

As Large Language Models (LLMs) are integrated into various sectors, ensuring their reliability and safety is crucial. This necessitates rigorous probing and auditing to maintain their effectiveness and trustworthiness in practical applications. Subjecting LLMs to varied iterations of a single query can unveil potential inconsistencies in their knowledge base or functional capacity. However, a tool for performing such audits with a easy to execute workflow, and low technical threshold is lacking. In this demo, we introduce ``AuditLLM,'' a novel tool designed to audit the performance of various LLMs in a methodical way. AuditLLM's primary function is to audit a given LLM by deploying multiple probes derived from a single question, thus detecting any inconsistencies in the model's comprehension or performance. A robust, reliable, and consistent LLM is expected to generate semantically similar responses to variably phrased versions of the same question. Building on this premise, AuditLLM generates easily interpretable results that reflect the LLM's consistency based on a single input question provided by the user. A certain level of inconsistency has been shown to be an indicator of potential bias, hallucinations, and other issues. One could then use the output of AuditLLM to further investigate issues with the aforementioned LLM. To facilitate demonstration and practical uses, AuditLLM offers two key modes: (1) Live mode which allows instant auditing of LLMs by analyzing responses to real-time queries; and (2) Batch mode which facilitates comprehensive LLM auditing by processing multiple queries at once for in-depth analysis. This tool is beneficial for both researchers and general users, as it enhances our understanding of LLMs' capabilities in generating responses, using a standardized auditing platform.

Bryan Guan, Tanya Roosta, Peyman Passban et al.

As large language models (LLMs) become integral to diverse applications, ensuring their reliability under varying input conditions is crucial. One key issue affecting this reliability is order sensitivity, wherein slight variations in the input arrangement can lead to inconsistent or biased outputs. Although recent advances have reduced this sensitivity, the problem remains unresolved. This paper investigates the extent of order sensitivity in LLMs whose internal components are hidden from users (such as closed-source models or those accessed via API calls). We conduct experiments across multiple tasks, including paraphrasing, relevance judgment, and multiple-choice questions. Our results show that input order significantly affects performance across tasks, with shuffled inputs leading to measurable declines in output accuracy. Few-shot prompting demonstrates mixed effectiveness and offers partial mitigation; however, fails to fully resolve the problem. These findings highlight persistent risks, particularly in high-stakes applications, and point to the need for more robust LLMs or improved input-handling techniques in future development.

Fatema Siddika, Md Anwar Hossen, J. Pablo Muñoz et al.

Parameter-efficient fine-tuning (PEFT) has attracted significant attention for adapting large pre-trained models by modifying a small subset of parameters. Recently, Representation Fine-tuning (ReFT) has emerged as an effective alternative. ReFT shifts the fine-tuning paradigm from updating model weights to directly manipulating hidden representations that capture rich semantic information, and performs better than state-of-the-art PEFTs in standalone settings. However, its application in Federated Learning (FL) remains challenging due to heterogeneity in clients' data distributions, model capacities, and computational resources. To address these challenges, we introduce Federated Representation Fine-Tuning (FedReFT), a novel approach to fine-tune the client's hidden representation. FedReFT applies sparse intervention layers to steer hidden representations directly, offering a lightweight and semantically rich fine-tuning alternative ideal for edge devices. However, representation-level updates are especially vulnerable to aggregation mismatch under different task heterogeneity, where naive averaging can corrupt semantic alignment. To mitigate this issue, we propose All-But-Me (ABM) aggregation, where each client receives the aggregated updates of others and partially incorporates them, enabling stable and personalized learning by balancing local focus with global knowledge. We evaluate FedReFT on commonsense reasoning, arithmetic reasoning, instruction-tuning, and GLUE, where it consistently outperforms state-of-the-art PEFT methods in FL, achieving 7x-15x higher parameter efficiency compared to leading LoRA-based approaches.

Julia Kharchenko, Tanya Roosta, Aman Chadha et al.

This paper introduces a comprehensive benchmark for evaluating how Large Language Models (LLMs) respond to linguistic shibboleths: subtle linguistic markers that can inadvertently reveal demographic attributes such as gender, social class, or regional background. Through carefully constructed interview simulations using 100 validated question-response pairs, we demonstrate how LLMs systematically penalize certain linguistic patterns, particularly hedging language, despite equivalent content quality. Our benchmark generates controlled linguistic variations that isolate specific phenomena while maintaining semantic equivalence, which enables the precise measurement of demographic bias in automated evaluation systems. We validate our approach along multiple linguistic dimensions, showing that hedged responses receive 25.6% lower ratings on average, and demonstrate the benchmark's effectiveness in identifying model-specific biases. This work establishes a foundational framework for detecting and measuring linguistic discrimination in AI systems, with broad applications to fairness in automated decision-making contexts.

Duy Phuong Nguyen, J. Pablo Munoz, Tanya Roosta et al.

Federated Learning (FL) enables collaborative learning across distributed clients while preserving data privacy. However, FL faces significant challenges when dealing with heterogeneous data distributions, which can lead to suboptimal global models that fail to generalize across diverse clients. In this work, we propose a novel framework designed to tackle these challenges by introducing a dual-adapter approach. The method utilizes a larger local adapter for client-specific personalization and a smaller global adapter to facilitate efficient knowledge sharing across clients. Additionally, we incorporate a pruning mechanism to reduce communication overhead by selectively removing less impactful parameters from the local adapter. Through extensive experiments on a range of vision and language tasks, our method demonstrates superior performance compared to existing approaches. It achieves higher test accuracy, lower performance variance among clients, and improved worst-case performance, all while significantly reducing communication and computation costs. Overall, the proposed method addresses the critical trade-off between model personalization and generalization, offering a scalable solution for real-world FL applications.

Preetam Prabhu Srikar Dammu, Arnav Palkhiwala, Tanya Roosta et al.

With the emergence of search-enabled generative QA systems, users are increasingly turning to tools that browse, aggregate, and reconcile evidence across multiple sources on their behalf. Yet many widely used QA benchmarks remain answerable by retrieving a single relevant passage, making them poorly suited for measuring cross-source sensemaking, such as integrating evidence, tracking causal links, and resolving dependencies across facets of a topic. We present iAgentBench, a dynamic ODQA benchmark that targets these higher-level information needs while keeping questions natural and grounded in realistic information-seeking behavior. iAgentBench draws seed topics from real-world attention signals and uses common user intent patterns to construct user-like questions whose answers require combining evidence from multiple sources, not just extracting a single snippet. Each instance is released with traceable evidence and auditable intermediate artifacts that support contamination checks and enable fine-grained diagnosis of failures in retrieval versus synthesis. Experiments across multiple LLMs show that retrieval improves accuracy, but retrieval alone does not reliably resolve these questions, underscoring the need to evaluate evidence use, not just evidence access.

Julia Kharchenko, Tanya Roosta, Aman Chadha et al.

Large Language Models (LLMs) attempt to imitate human behavior by responding to humans in a way that pleases them, including by adhering to their values. However, humans come from diverse cultures with different values. It is critical to understand whether LLMs showcase different values to the user based on the stereotypical values of a user's known country. We prompt different LLMs with a series of advice requests based on 5 Hofstede Cultural Dimensions -- a quantifiable way of representing the values of a country. Throughout each prompt, we incorporate personas representing 36 different countries and, separately, languages predominantly tied to each country to analyze the consistency in the LLMs' cultural understanding. Through our analysis of the responses, we found that LLMs can differentiate between one side of a value and another, as well as understand that countries have differing values, but will not always uphold the values when giving advice, and fail to understand the need to answer differently based on different cultural values. Rooted in these findings, we present recommendations for training value-aligned and culturally sensitive LLMs. More importantly, the methodology and the framework developed here can help further understand and mitigate culture and language alignment issues with LLMs.

Tanya Roosta, Peyman Passban, Ankit Chadha

Training neural machine translation (NMT) models in federated learning (FL) settings could be inefficient both computationally and communication-wise, due to the large size of translation engines as well as the multiple rounds of updates required to train clients and a central server. In this paper, we explore how to efficiently build NMT models in an FL setup by proposing a novel solution. In order to reduce the communication overhead, out of all neural layers we only exchange what we term "Controller" layers. Controllers are a small number of additional neural components connected to our pre-trained architectures. These new components are placed in between original layers. They act as liaisons to communicate with the central server and learn minimal information that is sufficient enough to update clients. We evaluated the performance of our models on five datasets from different domains to translate from German into English. We noted that the models equipped with Controllers preform on par with those trained in a central and non-FL setting. In addition, we observed a substantial reduction in the communication traffic of the FL pipeline, which is a direct consequence of using Controllers. Based on our experiments, Controller-based models are ~6 times less expensive than their other peers. This reduction is significantly important when we consider the number of parameters in large models and it becomes even more critical when such parameters need to be exchanged for multiple rounds in FL settings.