Ashish Kattamuri

Ashish Kattamuri


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
3 November 2025 in Preprints.org
Verified
Preprint
Open access
PDF file
AVArpita Vats
Arpita Vats
RRRahul Raja
Rahul Raja
AKAshish Kattamuri
Ashish Kattamuri
ABAbhinav Bohra
Abhinav Bohra
Quantum Natural Language Processing (QNLP) is an emerging interdisciplinary field at the intersection of quantum computing, natural language understanding, and formal linguistic theory. As advances in quantum hardware and algorithms accelerate, QNLP promises new paradigms for representation learning, semantic modeling, and efficient computation. However, existing literature remains fragmented, with no unified synthesis across modeling, encoding, and evaluation dimensions.In this work, we present the first systematic and taxonomy driven survey of QNLP that holistically organizes research spanning three core dimensions: computational models, encoding paradigms, and evaluation frameworks. First, we analyze foundational approaches that map linguistic structures into quantum formalism, including categorical compositional models, variational quantum circuits, and hybrid quantum classical architectures. Second, we introduce a unified taxonomy of encoding strategies, ranging from quantum tokenization and state preparation to embedding based encodings, highlighting tradeoffs in scalability, noise resilience, and expressiveness. Third, we provide the first comparative synthesis of evaluation methodologies, benchmark datasets, and performance metrics, while identifying reproducibility and standardization gaps.We further contrast quantum inspired NLP methods with fully quantum implemented systems, offering insights into resource efficiency, hardware feasibility, and real world applicability. Finally, we outline open challenges such as integration with LLMs and unified benchmark design, and propose a research agenda for advancing QNLP as a scalable and reliable discipline.

29 September 2025 in Preprints.org
Verified
Preprint
Open access
PDF file
AKAshish Kattamuri
Ashish Kattamuri
HFHarshwardhan Fartale
Harshwardhan Fartale
AVArpita Vats
Arpita Vats
RRRahul Raja
Rahul Raja
Data contamination poses a significant challenge to reliable LLM evaluation, where models may achieve high performance by memorizing training data rather than demonstrating genuine reasoning capabilities. We introduce RADAR (Recall vs. Reasoning Detection through Activation Representation), a novel framework that leverages mechanistic interpretability to detect contamination by distinguishing recall-based from reasoning-based model responses. RADAR extracts 37 features spanning surfacelevel confidence trajectories and deep mechanistic properties including attention specialization, circuit dynamics, and activation flow patterns. Using an ensemble of classifiers trained on these features, RADAR achieves 93% accuracy on a diverse evaluation set, with perfect performance on clear cases and 76.7% accuracy on challenging ambiguous examples. This work demonstrates the potential of mechanistic interpretability for advancing LLM evaluation beyond traditional surface-level metrics. The code used in this work is publicly available.

13 November 2025 in Preprints.org
Verified
Preprint
Open access
PDF file
AKAshish Kattamuri
Ashish Kattamuri
AVArpita Vats
Arpita Vats
HFHarshwardhan Fartale
Harshwardhan Fartale
RRRahul Raja
Rahul Raja
AMAkshata Kishore Moharir
Akshata Kishore Moharir
IPIshita Prasad
Ishita Prasad
Recursive prompting with large language models enables scalable synthetic dataset generation but introduces the risk of bias amplification. We investigate gender bias dynamics across three generations of recursive text generation using three complementary evaluation frameworks: rule-based pattern matching, embedding-based semantic similarity, and downstream task performance. Experiments with three initial bias levels (0.1, 0.3, 0.6) and four mitigation strategies reveal equilibrium dynamics rather than monotonic amplification. The low initial bias amplifies toward the model’s inherent bias level (+36%), whereas the high initial bias decays toward it (−26%). Among mitigation methods, contrastive augmentation, which introduces gender-swapped variants, achieves significant downstream bias reduction (98.8% for low initial bias and 91% on average) despite producing higher embedding-based bias scores. This paradox demonstrates that semantic similarity metrics may diverge from behavioral fairness outcomes, highlighting the need for multidimensional evaluation in responsible synthetic data generation.