Jwala Dhamala

Elan Markowitz, Anil Ramakrishna, Jwala Dhamala et al. · amazon-science

Knowledge graphs (KGs) complement Large Language Models (LLMs) by providing reliable, structured, domain-specific, and up-to-date external knowledge. However, KGs and LLMs are often developed separately and must be integrated after training. We introduce Tree-of-Traversals, a novel zero-shot reasoning algorithm that enables augmentation of black-box LLMs with one or more KGs. The algorithm equips a LLM with actions for interfacing a KG and enables the LLM to perform tree search over possible thoughts and actions to find high confidence reasoning paths. We evaluate on two popular benchmark datasets. Our results show that Tree-of-Traversals significantly improves performance on question answering and KG question answering tasks. Code is available at \url{https://github.com/amazon-science/tree-of-traversals}

Yang Trista Cao, Yada Pruksachatkun, Kai-Wei Chang et al. · amazon-science

Multiple metrics have been introduced to measure fairness in various natural language processing tasks. These metrics can be roughly categorized into two categories: 1) \emph{extrinsic metrics} for evaluating fairness in downstream applications and 2) \emph{intrinsic metrics} for estimating fairness in upstream contextualized language representation models. In this paper, we conduct an extensive correlation study between intrinsic and extrinsic metrics across bias notions using 19 contextualized language models. We find that intrinsic and extrinsic metrics do not necessarily correlate in their original setting, even when correcting for metric misalignments, noise in evaluation datasets, and confounding factors such as experiment configuration for extrinsic metrics. %al

Amazon AGI, Aaron Langford, Aayush Shah et al. · amazon-science

We present Amazon Nova, a new generation of state-of-the-art foundation models that deliver frontier intelligence and industry-leading price performance. Amazon Nova Pro is a highly-capable multimodal model with the best combination of accuracy, speed, and cost for a wide range of tasks. Amazon Nova Lite is a low-cost multimodal model that is lightning fast for processing images, video, documents and text. Amazon Nova Micro is a text-only model that delivers our lowest-latency responses at very low cost. Amazon Nova Canvas is an image generation model that creates professional grade images with rich customization controls. Amazon Nova Reel is a video generation model offering high-quality outputs, customization, and motion control. Our models were built responsibly and with a commitment to customer trust, security, and reliability. We report benchmarking results for core capabilities, agentic performance, long context, functional adaptation, runtime performance, and human evaluation.

Umang Gupta, Jwala Dhamala, Varun Kumar et al. · amazon-science, gatech

Language models excel at generating coherent text, and model compression techniques such as knowledge distillation have enabled their use in resource-constrained settings. However, these models can be biased in multiple ways, including the unfounded association of male and female genders with gender-neutral professions. Therefore, knowledge distillation without any fairness constraints may preserve or exaggerate the teacher model's biases onto the distilled model. To this end, we present a novel approach to mitigate gender disparity in text generation by learning a fair model during knowledge distillation. We propose two modifications to the base knowledge distillation based on counterfactual role reversal$\unicode{x2014}$modifying teacher probabilities and augmenting the training set. We evaluate gender polarity across professions in open-ended text generated from the resulting distilled and finetuned GPT$\unicode{x2012}$2 models and demonstrate a substantial reduction in gender disparity with only a minor compromise in utility. Finally, we observe that language models that reduce gender polarity in language generation do not improve embedding fairness or downstream classification fairness.

Ninareh Mehrabi, Palash Goyal, Apurv Verma et al. · amazon-science, gatech

Natural language often contains ambiguities that can lead to misinterpretation and miscommunication. While humans can handle ambiguities effectively by asking clarifying questions and/or relying on contextual cues and common-sense knowledge, resolving ambiguities can be notoriously hard for machines. In this work, we study ambiguities that arise in text-to-image generative models. We curate a benchmark dataset covering different types of ambiguities that occur in these systems. We then propose a framework to mitigate ambiguities in the prompts given to the systems by soliciting clarifications from the user. Through automatic and human evaluations, we show the effectiveness of our framework in generating more faithful images aligned with human intention in the presence of ambiguities.

Caleb Ziems, William Held, Jingfeng Yang et al. · amazon-science, gatech

Dialect differences caused by regional, social, and economic factors cause performance discrepancies for many groups of language technology users. Inclusive and equitable language technology must critically be dialect invariant, meaning that performance remains constant over dialectal shifts. Current systems often fall short of this ideal since they are designed and tested on a single dialect: Standard American English (SAE). We introduce a suite of resources for evaluating and achieving English dialect invariance. The resource is called Multi-VALUE, a controllable rule-based translation system spanning 50 English dialects and 189 unique linguistic features. Multi-VALUE maps SAE to synthetic forms of each dialect. First, we use this system to stress tests question answering, machine translation, and semantic parsing. Stress tests reveal significant performance disparities for leading models on non-standard dialects. Second, we use this system as a data augmentation technique to improve the dialect robustness of existing systems. Finally, we partner with native speakers of Chicano and Indian English to release new gold-standard variants of the popular CoQA task. To execute the transformation code, run model checkpoints, and download both synthetic and gold-standard dialectal benchmark datasets, see http://value-nlp.org.

Satyapriya Krishna, Rahul Gupta, Apurv Verma et al. · amazon-science, gatech

With the rapid growth in language processing applications, fairness has emerged as an important consideration in data-driven solutions. Although various fairness definitions have been explored in the recent literature, there is lack of consensus on which metrics most accurately reflect the fairness of a system. In this work, we propose a new formulation : ACCUMULATED PREDICTION SENSITIVITY, which measures fairness in machine learning models based on the model's prediction sensitivity to perturbations in input features. The metric attempts to quantify the extent to which a single prediction depends on a protected attribute, where the protected attribute encodes the membership status of an individual in a protected group. We show that the metric can be theoretically linked with a specific notion of group fairness (statistical parity) and individual fairness. It also correlates well with humans' perception of fairness. We conduct experiments on two text classification datasets : JIGSAW TOXICITY, and BIAS IN BIOS, and evaluate the correlations between metrics and manual annotations on whether the model produced a fair outcome. We observe that the proposed fairness metric based on prediction sensitivity is statistically significantly more correlated with human annotation than the existing counterfactual fairness metric.

Jwala Dhamala, Varun Kumar, Rahul Gupta et al. · amazon-science

Several prior works have shown that language models (LMs) can generate text containing harmful social biases and stereotypes. While decoding algorithms play a central role in determining properties of LM generated text, their impact on the fairness of the generations has not been studied. We present a systematic analysis of the impact of decoding algorithms on LM fairness, and analyze the trade-off between fairness, diversity and quality. Our experiments with top-$p$, top-$k$ and temperature decoding algorithms, in open-ended language generation, show that fairness across demographic groups changes significantly with change in decoding algorithm's hyper-parameters. Notably, decoding algorithms that output more diverse text also output more texts with negative sentiment and regard. We present several findings and provide recommendations on standardized reporting of decoding details in fairness evaluations and optimization of decoding algorithms for fairness alongside quality and diversity.

Ninareh Mehrabi, Palash Goyal, Anil Ramakrishna et al. · amazon-science

With the recent surge of language models in different applications, attention to safety and robustness of these models has gained significant importance. Here we introduce a joint framework in which we simultaneously probe and improve the robustness of a black-box target model via adversarial prompting and belief augmentation using iterative feedback loops. This framework utilizes an automated red teaming approach to probe the target model, along with a belief augmenter to generate instructions for the target model to improve its robustness to those adversarial probes. Importantly, the adversarial model and the belief generator leverage the feedback from past interactions to improve the effectiveness of the adversarial prompts and beliefs, respectively. In our experiments, we demonstrate that such a framework can reduce toxic content generation both in dynamic cases where an adversary directly interacts with a target model and static cases where we use a static benchmark dataset to evaluate our model.

Linyue Cai, Samuel Yeh, Jwala Dhamala et al.

Large language models can deceive by subtly manipulating truthful information -- omitting key facts, shifting focus, or obscuring meaning -- making such behavior difficult to detect. Existing black-box methods rely on coarse-grained judgments, offering limited interpretability and failing to pinpoint which facts were distorted and how. We introduce DECOR, a multi-agent framework grounded in Information Manipulation Theory for fine-grained auditing of strategic deception in LLM responses. DECOR decomposes input contexts into atomic informational units and scores each unit against the response across four dimensions of manipulation, producing interpretable manipulation profiles that are aggregated into a global deception index. We comprehensively evaluate DECOR on both single-turn and multi-turn deception detection benchmarks spanning real-world domains, and show that DECOR achieves state-of-the-art performance on both, outperforming competitive baselines. The framework generalizes across 15 frontier models, and ablation studies confirm the contribution of each key design component. Our findings demonstrate that fine-grained, theory-grounded auditing of information manipulation offers an effective and interpretable path for LLM deception detection.

Yang Xu, Xuanming Zhang, Samuel Yeh et al.

Deception is a pervasive feature of human communication and an emerging concern in large language models (LLMs). While recent studies document instances of LLM deception under pressure, most evaluations remain confined to single-turn prompts and fail to capture the long-horizon interactions in which deceptive strategies typically unfold. We introduce the first simulation framework for probing and evaluating deception in LLMs under extended sequences of interdependent tasks and dynamic contextual pressures. Our framework instantiates a multi-agent system: a performer agent tasked with completing tasks and a supervisor agent that evaluates progress, provides feedback, and maintains evolving states of trust. An independent deception auditor then reviews full trajectories to identify when and how deception occurs. We conduct extensive experiments across 11 frontier models, spanning both closed- and open-source systems, and find that deception is model-dependent, increases with event pressure, and consistently erodes supervisor trust. Qualitative analyses further reveal distinct strategies of concealment, equivocation, and falsification. Our findings establish deception as an emergent risk in long-horizon interactions and provide a foundation for evaluating future LLMs in real-world, trust-sensitive contexts.

Anaelia Ovalle, Ninareh Mehrabi, Palash Goyal et al. · amazon-science

Gender-inclusive NLP research has documented the harmful limitations of gender binary-centric large language models (LLM), such as the inability to correctly use gender-diverse English neopronouns (e.g., xe, zir, fae). While data scarcity is a known culprit, the precise mechanisms through which scarcity affects this behavior remain underexplored. We discover LLM misgendering is significantly influenced by Byte-Pair Encoding (BPE) tokenization, the tokenizer powering many popular LLMs. Unlike binary pronouns, BPE overfragments neopronouns, a direct consequence of data scarcity during tokenizer training. This disparate tokenization mirrors tokenizer limitations observed in multilingual and low-resource NLP, unlocking new misgendering mitigation strategies. We propose two techniques: (1) pronoun tokenization parity, a method to enforce consistent tokenization across gendered pronouns, and (2) utilizing pre-existing LLM pronoun knowledge to improve neopronoun proficiency. Our proposed methods outperform finetuning with standard BPE, improving neopronoun accuracy from 14.1% to 58.4%. Our paper is the first to link LLM misgendering to tokenization and deficient neopronoun grammar, indicating that LLMs unable to correctly treat neopronouns as pronouns are more prone to misgender.

Yuxuan Zhu, Tengjun Jin, Yada Pruksachatkun et al.

Benchmarks are essential for quantitatively tracking progress in AI. As AI agents become increasingly capable, researchers and practitioners have introduced agentic benchmarks to evaluate agents on complex, real-world tasks. These benchmarks typically measure agent capabilities by evaluating task outcomes via specific reward designs. However, we show that many agentic benchmarks have issues in task setup or reward design. For example, SWE-bench Verified uses insufficient test cases, while TAU-bench counts empty responses as successful. Such issues can lead to under- or overestimation of agents' performance by up to 100% in relative terms. To make agentic evaluation rigorous, we introduce the Agentic Benchmark Checklist (ABC), a set of guidelines that we synthesized from our benchmark-building experience, a survey of best practices, and previously reported issues. When applied to CVE-Bench, a benchmark with a particularly complex evaluation design, ABC reduces the performance overestimation by 33%.

Anaelia Ovalle, Palash Goyal, Jwala Dhamala et al.

Transgender and non-binary (TGNB) individuals disproportionately experience discrimination and exclusion from daily life. Given the recent popularity and adoption of language generation technologies, the potential to further marginalize this population only grows. Although a multitude of NLP fairness literature focuses on illuminating and addressing gender biases, assessing gender harms for TGNB identities requires understanding how such identities uniquely interact with societal gender norms and how they differ from gender binary-centric perspectives. Such measurement frameworks inherently require centering TGNB voices to help guide the alignment between gender-inclusive NLP and whom they are intended to serve. Towards this goal, we ground our work in the TGNB community and existing interdisciplinary literature to assess how the social reality surrounding experienced marginalization of TGNB persons contributes to and persists within Open Language Generation (OLG). This social knowledge serves as a guide for evaluating popular large language models (LLMs) on two key aspects: (1) misgendering and (2) harmful responses to gender disclosure. To do this, we introduce TANGO, a dataset of template-based real-world text curated from a TGNB-oriented community. We discover a dominance of binary gender norms reflected by the models; LLMs least misgendered subjects in generated text when triggered by prompts whose subjects used binary pronouns. Meanwhile, misgendering was most prevalent when triggering generation with singular they and neopronouns. When prompted with gender disclosures, TGNB disclosure generated the most stigmatizing language and scored most toxic, on average. Our findings warrant further research on how TGNB harms manifest in LLMs and serve as a broader case study toward concretely grounding the design of gender-inclusive AI in community voices and interdisciplinary literature.

Md Shakil Zaman, Jwala Dhamala, Pradeep Bajracharya et al.

Probabilistic estimation of cardiac electrophysiological model parameters serves an important step towards model personalization and uncertain quantification. The expensive computation associated with these model simulations, however, makes direct Markov Chain Monte Carlo (MCMC) sampling of the posterior probability density function (pdf) of model parameters computationally intensive. Approximated posterior pdfs resulting from replacing the simulation model with a computationally efficient surrogate, on the other hand, have seen limited accuracy. In this paper, we present a Bayesian active learning method to directly approximate the posterior pdf function of cardiac model parameters, in which we intelligently select training points to query the simulation model in order to learn the posterior pdf using a small number of samples. We integrate a generative model into Bayesian active learning to allow approximating posterior pdf of high-dimensional model parameters at the resolution of the cardiac mesh. We further introduce new acquisition functions to focus the selection of training points on better approximating the shape rather than the modes of the posterior pdf of interest. We evaluated the presented method in estimating tissue excitability in a 3D cardiac electrophysiological model in a range of synthetic and real-data experiments. We demonstrated its improved accuracy in approximating the posterior pdf compared to Bayesian active learning using regular acquisition functions, and substantially reduced computational cost in comparison to existing standard or accelerated MCMC sampling.

Yada Pruksachatkun, Satyapriya Krishna, Jwala Dhamala et al.

Existing bias mitigation methods to reduce disparities in model outcomes across cohorts have focused on data augmentation, debiasing model embeddings, or adding fairness-based optimization objectives during training. Separately, certified word substitution robustness methods have been developed to decrease the impact of spurious features and synonym substitutions on model predictions. While their end goals are different, they both aim to encourage models to make the same prediction for certain changes in the input. In this paper, we investigate the utility of certified word substitution robustness methods to improve equality of odds and equality of opportunity on multiple text classification tasks. We observe that certified robustness methods improve fairness, and using both robustness and bias mitigation methods in training results in an improvement in both fronts

Jwala Dhamala, Tony Sun, Varun Kumar et al.

Recent advances in deep learning techniques have enabled machines to generate cohesive open-ended text when prompted with a sequence of words as context. While these models now empower many downstream applications from conversation bots to automatic storytelling, they have been shown to generate texts that exhibit social biases. To systematically study and benchmark social biases in open-ended language generation, we introduce the Bias in Open-Ended Language Generation Dataset (BOLD), a large-scale dataset that consists of 23,679 English text generation prompts for bias benchmarking across five domains: profession, gender, race, religion, and political ideology. We also propose new automated metrics for toxicity, psycholinguistic norms, and text gender polarity to measure social biases in open-ended text generation from multiple angles. An examination of text generated from three popular language models reveals that the majority of these models exhibit a larger social bias than human-written Wikipedia text across all domains. With these results we highlight the need to benchmark biases in open-ended language generation and caution users of language generation models on downstream tasks to be cognizant of these embedded prejudices.

Ansel MacLaughlin, Jwala Dhamala, Anoop Kumar et al.

Neural Architecture Search (NAS) methods, which automatically learn entire neural model or individual neural cell architectures, have recently achieved competitive or state-of-the-art (SOTA) performance on variety of natural language processing and computer vision tasks, including language modeling, natural language inference, and image classification. In this work, we explore the applicability of a SOTA NAS algorithm, Efficient Neural Architecture Search (ENAS) (Pham et al., 2018) to two sentence pair tasks, paraphrase detection and semantic textual similarity. We use ENAS to perform a micro-level search and learn a task-optimized RNN cell architecture as a drop-in replacement for an LSTM. We explore the effectiveness of ENAS through experiments on three datasets (MRPC, SICK, STS-B), with two different models (ESIM, BiLSTM-Max), and two sets of embeddings (Glove, BERT). In contrast to prior work applying ENAS to NLP tasks, our results are mixed -- we find that ENAS architectures sometimes, but not always, outperform LSTMs and perform similarly to random architecture search.

Xiajun Jiang, Sandesh Ghimire, Jwala Dhamala et al.

Deep neural networks have shown great potential in image reconstruction problems in Euclidean space. However, many reconstruction problems involve imaging physics that are dependent on the underlying non-Euclidean geometry. In this paper, we present a new approach to learn inverse imaging that exploit the underlying geometry and physics. We first introduce a non-Euclidean encoding-decoding network that allows us to describe the unknown and measurement variables over their respective geometrical domains. We then learn the geometry-dependent physics in between the two domains by explicitly modeling it via a bipartite graph over the graphical embedding of the two geometry. We applied the presented network to reconstructing electrical activity on the heart surface from body-surface potential. In a series of generalization tasks with increasing difficulty, we demonstrated the improved ability of the presented network to generalize across geometrical changes underlying the data in comparison to its Euclidean alternatives.

Jwala Dhamala, John L. Sapp, B. Milan Horácek et al.

Estimation of patient-specific model parameters is important for personalized modeling, although sparse and noisy clinical data can introduce significant uncertainty in the estimated parameter values. This importance source of uncertainty, if left unquantified, will lead to unknown variability in model outputs that hinder their reliable adoptions. Probabilistic estimation model parameters, however, remains an unresolved challenge because standard Markov Chain Monte Carlo sampling requires repeated model simulations that are computationally infeasible. A common solution is to replace the simulation model with a computationally-efficient surrogate for a faster sampling. However, by sampling from an approximation of the exact posterior probability density function (pdf) of the parameters, the efficiency is gained at the expense of sampling accuracy. In this paper, we address this issue by integrating surrogate modeling into Metropolis Hasting (MH) sampling of the exact posterior pdfs to improve its acceptance rate. It is done by first quickly constructing a Gaussian process (GP) surrogate of the exact posterior pdfs using deterministic optimization. This efficient surrogate is then used to modify commonly-used proposal distributions in MH sampling such that only proposals accepted by the surrogate will be tested by the exact posterior pdf for acceptance/rejection, reducing unnecessary model simulations at unlikely candidates. Synthetic and real-data experiments using the presented method show a significant gain in computational efficiency without compromising the accuracy. In addition, insights into the non-identifiability and heterogeneity of tissue properties can be gained from the obtained posterior distributions.

Jwala Dhamala, Sandesh Ghimire, John L. Sapp et al.

The estimation of patient-specific tissue properties in the form of model parameters is important for personalized physiological models. However, these tissue properties are spatially varying across the underlying anatomical model, presenting a significance challenge of high-dimensional (HD) optimization at the presence of limited measurement data. A common solution to reduce the dimension of the parameter space is to explicitly partition the anatomical mesh, either into a fixed small number of segments or a multi-scale hierarchy. This anatomy-based reduction of parameter space presents a fundamental bottleneck to parameter estimation, resulting in solutions that are either too low in resolution to reflect tissue heterogeneity, or too high in dimension to be reliably estimated within feasible computation. In this paper, we present a novel concept that embeds a generative variational auto-encoder (VAE) into the objective function of Bayesian optimization, providing an implicit low-dimensional (LD) search space that represents the generative code of the HD spatially-varying tissue properties. In addition, the VAE-encoded knowledge about the generative code is further used to guide the exploration of the search space. The presented method is applied to estimating tissue excitability in a cardiac electrophysiological model. Synthetic and real-data experiments demonstrate its ability to improve the accuracy of parameter estimation with more than 10x gain in efficiency.

Jwala Dhamala, Sandesh Ghimire, John L. Sapp et al.

Personalization of cardiac models involves the optimization of organ tissue properties that vary spatially over the non-Euclidean geometry model of the heart. To represent the high-dimensional (HD) unknown of tissue properties, most existing works rely on a low-dimensional (LD) partitioning of the geometrical model. While this exploits the geometry of the heart, it is of limited expressiveness to allow partitioning that is small enough for effective optimization. Recently, a variational auto-encoder (VAE) was utilized as a more expressive generative model to embed the HD optimization into the LD latent space. Its Euclidean nature, however, neglects the rich geometrical information in the heart. In this paper, we present a novel graph convolutional VAE to allow generative modeling of non-Euclidean data, and utilize it to embed Bayesian optimization of large graphs into a small latent space. This approach bridges the gap of previous works by introducing an expressive generative model that is able to incorporate the knowledge of spatial proximity and hierarchical compositionality of the underlying geometry. It further allows transferring of the learned features across different geometries, which was not possible with a regular VAE. We demonstrate these benefits of the presented method in synthetic and real data experiments of estimating tissue excitability in a cardiac electrophysiological model.

Sandesh Ghimire, Jwala Dhamala, Prashnna Kumar Gyawali et al.

Noninvasive reconstruction of cardiac transmembrane potential (TMP) from surface electrocardiograms (ECG) involves an ill-posed inverse problem. Model-constrained regularization is powerful for incorporating rich physiological knowledge about spatiotemporal TMP dynamics. These models are controlled by high-dimensional physical parameters which, if fixed, can introduce model errors and reduce the accuracy of TMP reconstruction. Simultaneous adaptation of these parameters during TMP reconstruction, however, is difficult due to their high dimensionality. We introduce a novel model-constrained inference framework that replaces conventional physiological models with a deep generative model trained to generate TMP sequences from low-dimensional generative factors. Using a variational auto-encoder (VAE) with long short-term memory (LSTM) networks, we train the VAE decoder to learn the conditional likelihood of TMP, while the encoder learns the prior distribution of generative factors. These two components allow us to develop an efficient algorithm to simultaneously infer the generative factors and TMP signals from ECG data. Synthetic and real-data experiments demonstrate that the presented method significantly improve the accuracy of TMP reconstruction compared with methods constrained by conventional physiological models or without physiological constraints.

Sandesh Ghimire, Prashnna Kumar Gyawali, Jwala Dhamala et al.

Deep learning networks have shown state-of-the-art performance in many image reconstruction problems. However, it is not well understood what properties of representation and learning may improve the generalization ability of the network. In this paper, we propose that the generalization ability of an encoder-decoder network for inverse reconstruction can be improved in two means. First, drawing from analytical learning theory, we theoretically show that a stochastic latent space will improve the ability of a network to generalize to test data outside the training distribution. Second, following the information bottleneck principle, we show that a latent representation minimally informative of the input data will help a network generalize to unseen input variations that are irrelevant to the output reconstruction. Therefore, we present a sequence image reconstruction network optimized by a variational approximation of the information bottleneck principle with stochastic latent space. In the application setting of reconstructing the sequence of cardiac transmembrane potential from bodysurface potential, we assess the two types of generalization abilities of the presented network against its deterministic counterpart. The results demonstrate that the generalization ability of an inverse reconstruction network can be improved by stochasticity as well as the information bottleneck.

Jwala Dhamala, Emmanuel Azuh, Abdullah Al-Dujaili et al.

Timely prediction of clinically critical events in Intensive Care Unit (ICU) is important for improving care and survival rate. Most of the existing approaches are based on the application of various classification methods on explicitly extracted statistical features from vital signals. In this work, we propose to eliminate the high cost of engineering hand-crafted features from multivariate time-series of physiologic signals by learning their representation with a sequence-to-sequence auto-encoder. We then propose to hash the learned representations to enable signal similarity assessment for the prediction of critical events. We apply this methodological framework to predict Acute Hypotensive Episodes (AHE) on a large and diverse dataset of vital signal recordings. Experiments demonstrate the ability of the presented framework in accurately predicting an upcoming AHE.