Ayushi Agarwal

Ayushi Agarwal, Nisarg Patel, Neeraj Varshney et al. · amazon-science

Though state-of-the-art (SOTA) NLP systems have achieved remarkable performance on a variety of language understanding tasks, they primarily focus on questions that have a correct and a definitive answer. However, in real-world applications, users often ask questions that don't have a definitive answer. Incorrectly answering such questions certainly hampers a system's reliability and trustworthiness. Can SOTA models accurately identify such questions and provide a reasonable response? To investigate the above question, we introduce QnotA, a dataset consisting of five different categories of questions that don't have definitive answers. Furthermore, for each QnotA instance, we also provide a corresponding QA instance i.e. an alternate question that ''can be'' answered. With this data, we formulate three evaluation tasks that test a system's ability to 'identify', 'distinguish', and 'justify' QnotA questions. Through comprehensive experiments, we show that even SOTA models including GPT-3 and Flan T5 do not fare well on these tasks and lack considerably behind the human performance baseline. We conduct a thorough analysis which further leads to several interesting findings. Overall, we believe our work and findings will encourage and facilitate further research in this important area and help develop more robust models.

Ayushi Agarwal

We argue that formal certification of AI alignment over open-ended or unbounded input domains is impossible under standard assumptions in computational complexity and learning theory, and characterise what remains achievable. Two structurally independent impossibility theorems support this position. The semantic barrier (Theorem 1): deciding whether a system satisfies any non-trivial alignment property over the full input domain is NP-hard for feedforward networks and undecidable for Turing-complete architectures -- a direct consequence of neural-network verification complexity and Rice's Theorem. The statistical barrier (Theorem 2): any verification procedure that is both sound and tractable cannot satisfy Completeness over the full input domain -- a direct consequence of the impossibility of certifying infinite-domain properties from finite observations. These two theorems jointly entail a trilemma: no procedure can simultaneously satisfy soundness (no misaligned system is certified), completeness (no aligned system is rejected), and tractability (polynomial runtime). Each pair is simultaneously achievable; all three are not. We combine these results as a joint framework of two structurally independent barriers, prove their independence, and characterise the achievable Pareto frontier quantitatively via a constructive coverage-gap lower bound.

Sourav Banerjee, Ayushi Agarwal, Saloni Singla

As Large Language Models become more ubiquitous across domains, it becomes important to examine their inherent limitations critically. This work argues that hallucinations in language models are not just occasional errors but an inevitable feature of these systems. We demonstrate that hallucinations stem from the fundamental mathematical and logical structure of LLMs. It is, therefore, impossible to eliminate them through architectural improvements, dataset enhancements, or fact-checking mechanisms. Our analysis draws on computational theory and Godel's First Incompleteness Theorem, which references the undecidability of problems like the Halting, Emptiness, and Acceptance Problems. We demonstrate that every stage of the LLM process-from training data compilation to fact retrieval, intent classification, and text generation-will have a non-zero probability of producing hallucinations. This work introduces the concept of Structural Hallucination as an intrinsic nature of these systems. By establishing the mathematical certainty of hallucinations, we challenge the prevailing notion that they can be fully mitigated.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Ayushi Agarwal, Anannya Mathur, Preeti Ranjan Panda

The system-level cache is a critical resource shared by processor cores and domain-specific accelerators in heterogeneous systems on chips (SoCs). The strict QoS requirements of accelerators, such as deadlines, can lead to severe performance degradation of processor cores. Thus, managing the shared cache efficiently between cores and accelerators becomes crucial. State-of-the-art cache management techniques perform reuse-aware bypassing of accesses from cores with the help of reuse predictors to improve performance. However, architectural differences between accelerators and processor cores (often associated with deep cache hierarchies) can lead to significantly different reuse patterns at the shared cache. We propose a novel clustering-based methodology, LERN, for learning and predicting the reuse behavior of hardware accelerators at the shared cache. We then propose a deadline and reuse-aware cache management strategy, HyDRA, which explores a novel tradeoff between reuse and deadline awareness for performance efficiency. It uses LERN to dynamically predict the reuse behavior of the accelerator accesses and make bypass decisions to maximize the system throughput while meeting accelerator deadlines. We evaluate HyDRA across different workloads and varied accelerator configurations. It significantly improves the system performance and reduces the accelerator deadline miss rate.

Sourav Banerjee, Ayushi Agarwal, Eishkaran Singh

The pursuit of leaderboard rankings in Large Language Models (LLMs) has created a fundamental paradox: models excel at standardized tests while failing to demonstrate genuine language understanding and adaptability. Our systematic analysis of NLP evaluation frameworks reveals pervasive vulnerabilities across the evaluation spectrum, from basic metrics to complex benchmarks like GLUE and MMLU. These vulnerabilities manifest through benchmark exploitation, dataset contamination, and evaluation bias, creating a false perception of progress in language understanding capabilities. Through extensive review of contemporary evaluation approaches, we identify significant limitations in static benchmark designs, human evaluation protocols, and LLM-as-judge frameworks, all of which compromise the reliability of current performance assessments. As LLM capabilities evolve and existing benchmarks become redundant, we lay the groundwork for new evaluation methods that resist manipulation, minimize data contamination, and assess domain-specific tasks. This requires frameworks that are adapted dynamically, addressing current limitations and providing a more accurate reflection of LLM performance.

Sourav Banerjee, Ayushi Agarwal, Promila Ghosh

Automatic Speech Recognition (ASR) systems in the clinical domain face significant challenges, notably the need to recognise specialised medical vocabulary accurately and meet stringent precision requirements. We introduce United-MedASR, a novel architecture that addresses these challenges by integrating synthetic data generation, precision ASR fine-tuning, and advanced semantic enhancement techniques. United-MedASR constructs a specialised medical vocabulary by synthesising data from authoritative sources such as ICD-10 (International Classification of Diseases, 10th Revision), MIMS (Monthly Index of Medical Specialties), and FDA databases. This enriched vocabulary helps finetune the Whisper ASR model to better cater to clinical needs. To enhance processing speed, we incorporate Faster Whisper, ensuring streamlined and high-speed ASR performance. Additionally, we employ a customised BART-based semantic enhancer to handle intricate medical terminology, thereby increasing accuracy efficiently. Our layered approach establishes new benchmarks in ASR performance, achieving a Word Error Rate (WER) of 0.985% on LibriSpeech test-clean, 0.26% on Europarl-ASR EN Guest-test, and demonstrating robust performance on Tedlium (0.29% WER) and FLEURS (0.336% WER). Furthermore, we present an adaptable architecture that can be replicated across different domains, making it a versatile solution for domain-specific ASR systems.

Sourav Banerjee, Anush Mahajan, Ayushi Agarwal et al.

Natural Language Inference (NLI) tasks require identifying the relationship between sentence pairs, typically classified as entailment, contradiction, or neutrality. While the current state-of-the-art (SOTA) model, Entailment Few-Shot Learning (EFL), achieves a 93.1% accuracy on the Stanford Natural Language Inference (SNLI) dataset, further advancements are constrained by the dataset's limitations. To address this, we propose a novel approach leveraging synthetic data augmentation to enhance dataset diversity and complexity. We present UnitedSynT5, an advanced extension of EFL that leverages a T5-based generator to synthesize additional premise-hypothesis pairs, which are rigorously cleaned and integrated into the training data. These augmented examples are processed within the EFL framework, embedding labels directly into hypotheses for consistency. We train a GTR-T5-XL model on this expanded dataset, achieving a new benchmark of 94.7% accuracy on the SNLI dataset, 94.0% accuracy on the E-SNLI dataset, and 92.6% accuracy on the MultiNLI dataset, surpassing the previous SOTA models. This research demonstrates the potential of synthetic data augmentation in improving NLI models, offering a path forward for further advancements in natural language understanding tasks.

Divyagna Bavikadi, Ayushi Agarwal, Shashank Ganta et al.

Recent advances in experimental methods have enabled researchers to collect data on thousands of analytes simultaneously. This has led to correlational studies that associated molecular measurements with diseases such as Alzheimer's, Liver, and Gastric Cancer. However, the use of thousands of biomarkers selected from the analytes is not practical for real-world medical diagnosis and is likely undesirable due to potentially formed spurious correlations. In this study, we evaluate 4 different methods for biomarker selection and 4 different machine learning (ML) classifiers for identifying correlations, evaluating 16 approaches in all. We found that contemporary methods outperform previously reported logistic regression in cases where 3 and 10 biomarkers are permitted. When specificity is fixed at 0.9, ML approaches produced a sensitivity of 0.240 (3 biomarkers) and 0.520 (10 biomarkers), while standard logistic regression provided a sensitivity of 0.000 (3 biomarkers) and 0.040 (10 biomarkers). We also noted that causal-based methods for biomarker selection proved to be the most performant when fewer biomarkers were permitted, while univariate feature selection was the most performant when a greater number of biomarkers were permitted.