arXiv subjects

Kanak Raj

Publications and source records attributed to Kanak Raj.

6 recordsLinked to original sources

arXiv2026

Test-Time Scaling in the Wild: Why Exploitation, Not Exploration, Is the Bottleneck

Test-time scaling (TTS) improves language model outputs by spending additional inference compute - generating multiple candidates, searching over partial sequences, or iteratively refining drafts. These techniques yield large gains on mathematics and code, but have been developed and stress-tested almost exclusively on tasks where verification is straightforward. We conduct the first compute-normalised comparison of five TTS families across five open-ended generation benchmarks spanning medicine, law, finance, general chat, and creative writing - grounded in a unified framework that decomposes the effectiveness of each method's token budget into exploration and exploitation. The answer depends on which side of that decomposition you examine. Scaling exploration works: the best candidate in the pool improves steadily with compute across all settings. What breaks is exploitation - the step that converts a rich candidate pool into a final output. With state-of-the-art generators, reward models correlate at only $\rho_v \approx 0.12$ with true quality, rendering selection near-random regardless of budget. Tree search amplifies this failure through diversity collapse. Refinement helps on one of five benchmarks; its apparent gains elsewhere are confounded. Only synthesis across candidates (Fusion) consistently improves over single-sample baselines, yet still recovers only ~40% of available quality. The candidate pool is not the bottleneck - choosing from it is.

cs.CL↗

arXiv2026

LLMs Remember First, Forget Last: Dual-Process Interference in Large Language Models

Large language models can process millions of tokens, yet how they handle conflicting information within context remains poorly understood. From patient health logs tracking evolving vital signs to legal documents with superseding clauses, real-world applications routinely require models to retrieve specific values from streams of semantically similar, competing updates. We adapt classical interference paradigms from cognitive psychology to compare retroactive interference (RI; recalling initial values after updates) and proactive interference (PI; recalling recent values despite competing prior encodings) across 39 LLMs spanning 1B to 2.5T parameters. Every model exhibits the same pattern: PI causes substantially greater degradation than RI (d = 1.73), the opposite of the typical human finding where new information more readily disrupts old. Four lines of evidence indicate that RI and PI engage distinct mechanisms: model size predicts RI resistance but not PI; the two scores correlate only weakly; error analysis reveals qualitatively different failure profiles, with RI failures reflecting passive retrieval failure and PI failures reflecting active primacy intrusion; and hallucination rates remain below 1% in both conditions. An adaptive bootstrap design further shows that RI decays smoothly while PI collapses bimodally at high interference, consistent with gradual dilution versus winner-take-all attention dynamics. These findings characterize a systematic primacy bias in transformer attention, with implications for reliable deployment in settings where in-context information evolves over time.

cs.IR↗

arXiv2025

Thermal Degradation Mechanisms and Stability Enhancement Strategies in Perovskite Solar Cells: A Review

Perovskite Solar Cells (PSCs) have garnered global research interest owing to their superior photovoltaic (PV) performance. The future of photovoltaic technology lies in PSCs since they can produce power with performance on par with the best silicon solar cells while being less expensive. PSCs have enormous potential; in just ten years, their efficiency increased from 3.8% to 25.2%, and research into new developments is still ongoing. Thermal instability is PSCs' main disadvantage, despite their high efficiency, flexibility, and lightweight nature. This paper looks at how temperature affects the ways that hole transport layers (HTLs) like spiro-OMeTAD and perovskite layers, especially MAPbI3, degrade. Elevated temperatures cause MAPbI3 to degrade into PbI2, CH3I, and NH3, with decomposition rates affected by moisture, oxygen, and environmental factors. Mixed cation compositions, such as Cs-MA-FA, have higher thermal stability, whereas MA+ cations break-down faster under heat stress. HTLs deteriorate due to morphological changes and the hydrophilicity of dopant additions like Li-TFSI and t-BP. Alternative dopant-free HTMs, such as P3HT and inorganic materials including CuSCN, NiOx, and Cu2O, have shown improved thermal stability and efficiency. Hybrid HTLs, dopant-free designs, and interface tweaks are all viable solutions for increasing the stability of PSC. Addressing thermal stability issues remains crucial for the development of more reliable and efficient PSC technology.

cond-mat.mtrl-sci↗

arXiv2025

Numerical Investigations of Jet A Hexane Binary Fuel Droplet Impact on a Heated Solid Surface

In the present work, Jet A-Hexane binary fuel droplet impact dynamics on heated solid surfaces were studied numerically. This study is crucial for practical applications such as fuel injection in combustors and thermal management of engine components. Volume of fluid (VOF) method was used to analyse the impact dynamics, spreading behaviour, vaporisation, and heat transfer of n-hexane and Jet-A blended fuel droplets on heated stainless-steel surfaces. Droplet impact dynamics were investigated for two Weber numbers, i.e., 25 and 50, and surface temperatures ranging from 50C to 227C to capture transitions from gentle spreading to nucleate boiling and rebound phenomena. This work examines how fuel blending influences inertia, lamella formation, vapour recoil, and film boiling regimes. The results show that higher inertia in blended fuels enhances spreading but also triggers stronger vapour recoil at elevated temperatures, leading to droplet rebound. In contrast, pure hexane transitions to a stable film boiling regime at high surface temperatures, resulting in a decline in smoother heat flux. New correlations were developed linking Weber number, spreading ratio, and wall heat flux, offering predictive insights for real-world combustion scenarios. These findings advance the understanding of bi-component fuel droplet impacts on heated surfaces and provide a framework for designing efficient spray systems in combustors and thermal management in propulsion and power generation applications.

physics.flu-dyn↗

arXiv2024

K-PERM: Personalized Response Generation Using Dynamic Knowledge Retrieval and Persona-Adaptive Queries

Personalizing conversational agents can enhance the quality of conversations and increase user engagement. However, they often lack external knowledge to appropriately tend to a user's persona. This is particularly crucial for practical applications like mental health support, nutrition planning, culturally sensitive conversations, or reducing toxic behavior in conversational agents. To enhance the relevance and comprehensiveness of personalized responses, we propose using a two-step approach that involves (1) selectively integrating user personas and (2) contextualizing the response with supplementing information from a background knowledge source. We develop K-PERM (Knowledge-guided PErsonalization with Reward Modulation), a dynamic conversational agent that combines these elements. K-PERM achieves state-of-the-art performance on the popular FoCus dataset, containing real-world personalized conversations concerning global landmarks. We show that using responses from K-PERM can improve performance in state-of-the-art LLMs (GPT 3.5) by 10.5%, highlighting the impact of K-PERM for personalizing chatbots.

cs.IR↗

arXiv2023

Cook-Gen: Robust Generative Modeling of Cooking Actions from Recipes

As people become more aware of their food choices, food computation models have become increasingly popular in assisting people in maintaining healthy eating habits. For example, food recommendation systems analyze recipe instructions to assess nutritional contents and provide recipe recommendations. The recent and remarkable successes of generative AI methods, such as auto-regressive large language models, can lead to robust methods for a more comprehensive understanding of recipes for healthy food recommendations beyond surface-level nutrition content assessments. In this study, we explore the use of generative AI methods to extend current food computation models, primarily involving the analysis of nutrition and ingredients, to also incorporate cooking actions (e.g., add salt, fry the meat, boil the vegetables, etc.). Cooking actions are notoriously hard to model using statistical learning methods due to irregular data patterns - significantly varying natural language descriptions for the same action (e.g., marinate the meat vs. marinate the meat and leave overnight) and infrequently occurring patterns (e.g., add salt occurs far more frequently than marinating the meat). The prototypical approach to handling irregular data patterns is to increase the volume of data that the model ingests by orders of magnitude. Unfortunately, in the cooking domain, these problems are further compounded with larger data volumes presenting a unique challenge that is not easily handled by simply scaling up. In this work, we propose novel aggregation-based generative AI methods, Cook-Gen, that reliably generate cooking actions from recipes, despite difficulties with irregular data patterns, while also outperforming Large Language Models and other strong baselines.

cs.CL↗