Searcharxiv⌕ Search

arXiv subjects

Pranav Sharma

Publications and source records attributed to Pranav Sharma.

At least 19 recordsLinked to original sources

Advancing Quantization for R&R Models with Microscaling Formats

Recommendation and Ranking (R&R) models, particularly in digital advertising, require massive inference throughput but exhibit extreme sensitivity to precision loss. Unlike Large Language Models (LLMs) that frequently tolerate quantization-induced accuracy drops of 1% or more, production-grade R&R models enforce a stringent launch bar: model quality must remain within a 0.01% Normalized Entropy (NE) margin relative to the baseline, which commonly uses FP16 or FP32. This tight constraint has historically hindered the adoption of aggressive quantization in this domain. In this paper, we present an empirical evaluation of both MXFP8 and the 6-bit MXFP6 format for industrial-scale R&R inference on the AMD MI350X architecture. To the best of our knowledge, this is the first paper, across R&R and LLM domains, that reports significant speedups from native MXFP6 GEMMs, which quantize both weights and activations and compute directly on 6-bit elements, without sacrificing model quality. We demonstrate that we can selectively apply this 6-bit format to maximize overall QPS gain without breaching the strict NE parity threshold. We further show that MXFP8 is a viable alternative to row-wise FP8: the two formats have nearly the same accuracy on our models, so the faster kernel can be chosen for each GEMM shape. On two production ads ranking models, our implementation shows measured GEMM kernel speedups over FP16 of 1.33--1.55x for MXFP8 and 1.73--1.99x for MXFP6 across all supported linear layers, and up to 2.90x for MXFP6 on individual layers, alongside end-to-end (E2E) inference throughput gains from selective MXFP6 of 1.5--5.3% for Model A and 7.6--14.3% for Model B over their FP8 configurations.

cs.AI↗

BRST quantization for the restoration of broken symmetries: a pedagogical example

We present a pedagogical showcase of BRST quantization for restoring symmetries in nuclear many-body systems as a reformulation and potential alternative to conventional projection methods. The formalism is illustrated using translational invariance for a simple system of two interacting masses in one dimension, but with an eye toward generalizing to more particles, higher dimensions, and other symmetries. We explain the considerations underlying particular choices within the BRST construction, and develop both Hamiltonian and path integral formulations to provide guidance for the variety of many-body and effective field theory contexts where gauge fixing for symmetry restoration might be useful. For the demonstration system we show how to diagonalize within the extended BRST phase space, how variation after projection is recovered for product reference states, how the corresponding gauge-fixed functional integral is constructed, and how collective zero modes are isolated and controlled. Throughout, we keep in mind extensions of BRST symmetry to various approximation schemes as a guide for consistent symmetry restoration.

nucl-th↗

IndicFDB: Benchmarking Full-Duplex Voice Agents across Indian Languages

Full-duplex voice agents must handle pauses, take turns, backchannel, and respond to user interruptions in real time. Full-Duplex-Bench evaluates these behaviors, but its English-only corpus and reliance on word-timestamped ASR and an English-prompted LLM judge make it difficult to extend to Indian languages. We introduce IndicFDB, which extends it to ten languages spoken in India with 12,350 samples, nearly 17 times as many as the original. We address three challenges: finding conversational events in multilingual speech, evaluating their timing without reliable word-level alignment, and judging responses across languages. We mine pause handling, turn taking, and backchanneling samples from roughly 50,000 hours of channel-separated conversations using voice activity detection (VAD), and construct human-validated synthetic user interruption samples. Language-independent VAD heuristics evaluate timing, while an open-weight transcription and translation pipeline converts responses to English for LLM ratings of relevance and quality. Across seven voice agents, commercial APIs show unexpectedly consistent behavior across languages but are either fast or robust to pauses, never both, while monolingual open full-duplex models expose further tradeoffs among backchanneling, response quality, and latency.

cs.CL↗

Human-1 by Josh Talks: A Full-Duplex Conversational Modeling Framework in Hindi using Real-World Conversations

Full-duplex spoken dialogue systems can model natural conversational behaviours such as interruptions, overlaps, and backchannels, yet such systems remain largely unexplored for Indian languages. We present the first open, reproducible full-duplex spoken dialogue system for Hindi by adapting Moshi, a state-of-the-art duplex speech architecture, using a custom Hindi tokeniser and training on 26,000 hours of real spontaneous conversations collected from 14,695 speakers with separate speaker channels, enabling direct learning of turn-taking and overlap patterns from natural interactions. To support Hindi text generation, we replace the original English tokeniser and reinitialise text-vocabulary-dependent parameters while retaining the pre-trained audio components. We propose a two-stage training recipe -- large-scale pre-training followed by fine-tuning on 1,000 hours of conversational data. Evaluation through the prompted dialogue continuation paradigm with both automatic metrics and human judgments demonstrates that the resulting model generates natural and meaningful full-duplex conversational behaviour in Hindi. This work serves as a first step toward real-time duplex spoken dialogue systems for Hindi and other Indian languages.

cs.CL↗

Mirage: Unveiling Hidden Artifacts in Synthetic Images with Large Vision-Language Models

Recent advances in image generation models have led to models that produce synthetic images that are increasingly difficult for standard AI detectors to identify, even though they often remain distinguishable by humans. To identify this discrepancy, we introduce \textbf{Mirage}, a curated dataset comprising a diverse range of AI-generated images exhibiting visible artifacts, where current state-of-the-art detection methods largely fail. Furthermore, we investigate whether Large Vision-Language Models (LVLMs), which are increasingly employed as substitutes for human judgment in various tasks, can be leveraged for explainable AI image detection. Our experiments on both Mirage and existing benchmark datasets demonstrate that while LVLMs are highly effective at detecting AI-generated images with visible artifacts, their performance declines when confronted with images lacking such cues.

cs.CV↗

Testing Variational Perturbation Theory for Effective Actions Using the Gaudin-Yang Model

The background field formalism based on effective actions is a compelling framework for developing an effective field theory for nuclear density functional theory. Among the challenges in carrying out this development is handling both the particle-hole and pairing channels beyond the mean-field level, which includes how to incorporate collective degrees of freedom. Here we use the exactly solvable one-dimensional Gaudin-Yang model as a theoretical laboratory to explore candidate approaches. We compare Variational Perturbation Theory (VPT) to ordinary many-body perturbation theory and the inversion method, all to second order in their respective expansions, and verify issues with Hubbard-Stratonovich auxiliary fields. VPT outperforms the other approaches at this level over a wide range of densities. The next steps to extend this approach toward nuclei are outlined.

nucl-th↗

PersonaFlow: Designing LLM-Simulated Expert Perspectives for Enhanced Research Ideation

Generating interdisciplinary research ideas requires diverse domain expertise, but access to timely feedback is often limited by the availability of experts. In this paper, we introduce PersonaFlow, a novel system designed to provide multiple perspectives by using LLMs to simulate domain-specific experts. Our user studies showed that the new design 1) increased the perceived relevance and creativity of ideated research directions, and 2) promoted users' critical thinking activities (e.g., interpretation, analysis, evaluation, inference, and self-regulation), without increasing their perceived cognitive load. Moreover, users' ability to customize expert profiles significantly improved their sense of agency, which can potentially mitigate their over-reliance on AI. This work contributes to the design of intelligent systems that augment creativity and collaboration, and provides design implications of using customizable AI-simulated personas in domains within and beyond research ideation.

cs.HC↗

EvAlignUX: Advancing UX Evaluation through LLM-Supported Metrics Exploration

Evaluating UX in the context of AI's complexity, unpredictability, and generative nature presents unique challenges. How can we support HCI researchers to create comprehensive UX evaluation plans? In this paper, we introduce EvAlignUX, a system powered by large language models and grounded in scientific literature, designed to help HCI researchers explore evaluation metrics and their relationship to research outcomes. A user study with 19 HCI scholars showed that EvAlignUX improved the perceived quality and confidence in UX evaluation plans while prompting deeper consideration of research impact and risks. The system enhanced participants' thought processes, leading to the creation of a ``UX Question Bank'' to guide UX evaluation development. Findings also highlight how researchers' backgrounds influence their inspiration and concerns about AI over-reliance, pointing to future research on AI's role in fostering critical thinking. In a world where experience defines impact, we discuss the importance of shifting UX evaluation from a ``method-centric'' to a ``mindset-centric'' approach as the key to meaningful and lasting design evaluation.

cs.HC↗

Passive Snapshot Coded Aperture Dual-Pixel RGB-D Imaging

Passive, compact, single-shot 3D sensing is useful in many application areas such as microscopy, medical imaging, surgical navigation, and autonomous driving where form factor, time, and power constraints can exist. Obtaining RGB-D scene information over a short imaging distance, in an ultra-compact form factor, and in a passive, snapshot manner is challenging. Dual-pixel (DP) sensors are a potential solution to achieve the same. DP sensors collect light rays from two different halves of the lens in two interleaved pixel arrays, thus capturing two slightly different views of the scene, like a stereo camera system. However, imaging with a DP sensor implies that the defocus blur size is directly proportional to the disparity seen between the views. This creates a trade-off between disparity estimation vs. deblurring accuracy. To improve this trade-off effect, we propose CADS (Coded Aperture Dual-Pixel Sensing), in which we use a coded aperture in the imaging lens along with a DP sensor. In our approach, we jointly learn an optimal coded pattern and the reconstruction algorithm in an end-to-end optimization setting. Our resulting CADS imaging system demonstrates improvement of >1.5dB PSNR in all-in-focus (AIF) estimates and 5-6% in depth estimation quality over naive DP sensing for a wide range of aperture settings. Furthermore, we build the proposed CADS prototypes for DSLR photography settings and in an endoscope and a dermoscope form factor. Our novel coded dual-pixel sensing approach demonstrates accurate RGB-D reconstruction results in simulations and real-world experiments in a passive, snapshot, and compact manner.

eess.IV↗

Global Data in Astronomy: Challenges and Opportunities

Policy Brief on "Global Data in Astronomy: Challenges and Opportunities", distilled from the corresponding panel that was part of the discussions during S20 Policy Webinar on Astroinformatics for Sustainable Development held on 6-7 July 2023. Astronomy is increasingly becoming a data-driven science. Advances in our understanding of the physical mechanisms at work in the Universe require building ever-more sensitive telescopes to gather observations of the cosmos to test and advance our theoretical models of how the universe works. To confront the observed data with our theoretical models we require data hosting, archiving and storage and high-performance computing resources to run the theoretical calculations and compare our simulated and observed universe. We also require the sophisticated development of highly skilled human resources. Newer large projects are often run through international collaborations and partnerships, driving a need for 'open science' and collaborative structure across national boundaries. While astronomical data are useful scientifically, the data do not come with the same ethical/privacy-related restrictions as medical/biological data. Moreover, the ability to use data for new scientific analysis extends and expands the impact and reach of scientific surveys -- this is a strength that national funding agencies should capitalize on. We discuss the management and analysis of such large volumes of data and the corresponding significant challenges that require policy-level preparations. The policy webinar took place during the G20 presidency in India (2023). A summary based on the seven panels can be found here: arxiv:2401.04623.

astro-ph.IM↗

Long Term Space Data and Informatics Needs

Policy Brief on "Long Term Space Data and Informatics Needs", distilled from the corresponding panel that was part of the discussions during S20 Policy Webinar on Astroinformatics for Sustainable Development held on 6-7 July 2023. Persistent space data gathering, retention, transmission, and analysis play a pivotal role in deepening our grasp of the Universe and fostering the achievement of global sustainable development goals. Long-term data storage and curation is crucial not only to make the wide range of burgeoning data sets available to the global science community, but also to stabilize those data sets, enabling new science in the future to analyse long-term trends over unprecedented time spans. In addition to this, over the long-term, the imperative to store all data on the ground should be ameliorated by use of space-based data stores --maintained and seen to be as reliable as any other data archive. This concept is sometimes referred to as Memory of the Sky. Storing the data must be accompanied by the ability to analyse them. Several concepts covered below acknowledge roots and inspiration based in the Virtual Observatory effort. Within this policy document, we delve into the complexities surrounding the long-term utilization of space data and informatics, shedding light on the challenges and opportunities inherent in this endeavour. Further, we present a series of pragmatic recommendations designed to address these challenges proactively. The policy webinar took place during the G20 presidency in India (2023). A summary based on the seven panels can be found here: arxiv:2401.04623.

astro-ph.IM↗

Latest Developments and Opportunities in Sky Survey

Policy Brief on "Latest Developments and Opportunities in Sky Survey", distilled from the corresponding panel that was part of the discussions during S20 Policy Webinar on Astroinformatics for Sustainable Development held on 6-7 July 2023. Sky surveys have been a crucial tool in advancing our understanding of the Universe. The last few decades have seen an explosion in the number and scope of sky surveys, both ground-based and space-based. This growth has led to a wealth of data that has enabled us to make significant advances in many areas of astronomy, and help understand the physics of the universe. They have helped us discover new astronomical objects, the origin of the elements, dark matter and dark energy, the accelerated expansion of the universe, and gravitational waves. They have helped us study the distribution of neutral and ionized matter in the Universe and test our theories about the origin and evolution of galaxies, stars, and planets. We explore recent advances and potential avenues in sky surveys, and examine how these developments may impact the field of international astronomical research. The policy webinar took place during the G20 presidency in India (2023). A summary based on the seven panels can be found here: arxiv:2401.04623.

astro-ph.IM↗

Workforce Development in Astronomy and Astroinformatics

Policy Brief on "Workforce Development in Astronomy and Astroinformatics", distilled from the corresponding panel that was part of the discussions during S20 Policy Webinar on Astroinformatics for Sustainable Development held on 6-7 July 2023. The discipline of astronomy and astroinformatics is dynamically evolving thereby creating a compelling opportunity to foster a more inclusive, diverse, and proficient workforce. This is crucial for addressing multifaceted challenges that emerge as we progress and harness the potential therein. To realize this goal, it's imperative to cultivate strategies that promote inclusive practices in STEM education, encourage participation from historically excluded groups, provide training and mentorship, as well as provide active champions, especially for students and early career professionals from (historically) excluded groups. We provide an overview of the current status, resources available, and possible steps especially keeping in mind large international projects. The policy webinar took place during the G20 presidency in India (2023). A summary based on the seven panels can be found here: arxiv:2401.04623.

astro-ph.IM↗

Large Projects in Astronomy: An Indian Endeavour

Policy Brief on "Large Projects in Astronomy: An Indian Endeavour", distilled from the corresponding panel that was part of the discussions during S20 Policy Webinar on Astroinformatics for Sustainable Development held on 6-7 July 2023. Cutting-edge astronomy initiatives often entail substantial investment and require a high level of expertise. Even the most technologically advanced nations recognize the value of establishing international partnerships to secure both financial resources and talent for these ambitious endeavours as they hold immense promise for catalysing transformative scientific discoveries, driving technological innovation, provide training opportunities for the next generation of scientists and engineers, and expanding our understanding of the cosmos that surrounds us. Crucially, large-scale multilateral collaborations serve as powerful agents for promoting unity and peace among the global population. Participants from various nations share a vested interest in the success of these projects and the wealth of knowledge they yield, fostering a sense of common purpose and shared goals. By utilizing astroinformatics capabilities, these initiatives are not merely enhancing our comprehension of the universe but are also actively contributing to the attainment of sustainable development objectives. In this discussion, we delve into the challenges faced, and prospects for substantial astronomical undertakings. Additionally, we present recommendations aimed at guaranteeing their effectiveness and optimizing their influence on both scientific advancement and society. The policy webinar took place during the G20 presidency in India (2023). A summary based on the seven panels can be found here: arxiv:2401.04623.

astro-ph.IM↗

Regional and Global Collaborations in Astronomy

Policy Brief on "Regional and Global Collaborations in Astronomy", distilled from the corresponding panel that was part of the discussions during S20 Policy Webinar on Astroinformatics for Sustainable Development held on 6-7 July 2023. Astronomy brings together advanced scientific research, state-of-the-art technology, and educational initiatives, all while captivating and stimulating people of all ages. By doing so, it possesses the potential to serve as a powerful catalyst for sustainable global development and the resolution of global societal issues. It attracts a diverse range of scientists and experts from various fields, fostering collaboration and innovation. By leveraging their resources, influence, and diplomatic initiatives, S20 academies can foster an enabling environment for international collaborations in astronomy, facilitate knowledge exchange, and drive scientific advancements that benefit humanity. This policy brief explores the opportunities and challenges presented by regional and global collaborations in astronomy. The policy webinar took place during the G20 presidency in India (2023). A summary based on the seven panels can be found here: arxiv:2401.04623.

astro-ph.IM↗

AstroInformatics: Recommendations for Global Cooperation

Policy Brief on "AstroInformatics, Recommendations for Global Collaboration", distilled from panel discussions during S20 Policy Webinar on Astroinformatics for Sustainable Development held on 6-7 July 2023. The deliberations encompassed a wide array of topics, including broad astroinformatics, sky surveys, large-scale international initiatives, global data repositories, space-related data, regional and international collaborative efforts, as well as workforce development within the field. These discussions comprehensively addressed the current status, notable achievements, and the manifold challenges that the field of astroinformatics currently confronts. The G20 nations present a unique opportunity due to their abundant human and technological capabilities, coupled with their widespread geographical representation. Leveraging these strengths, significant strides can be made in various domains. These include, but are not limited to, the advancement of STEM education and workforce development, the promotion of equitable resource utilization, and contributions to fields such as Earth Science and Climate Science. We present a concise overview, followed by specific recommendations that pertain to both ground-based and space data initiatives. Our team remains readily available to furnish further elaboration on any of these proposals as required. Furthermore, we anticipate further engagement during the upcoming G20 presidencies in Brazil (2024) and South Africa (2025) to ensure the continued discussion and realization of these objectives. The policy webinar took place during the G20 presidency in India (2023). Notes based on the seven panels will be separately published.

astro-ph.IM↗

Achieving RGB-D level Segmentation Performance from a Single ToF Camera

Depth is a very important modality in computer vision, typically used as complementary information to RGB, provided by RGB-D cameras. In this work, we show that it is possible to obtain the same level of accuracy as RGB-D cameras on a semantic segmentation task using infrared (IR) and depth images from a single Time-of-Flight (ToF) camera. In order to fuse the IR and depth modalities of the ToF camera, we introduce a method utilizing depth-specific convolutions in a multi-task learning framework. In our evaluation on an in-car segmentation dataset, we demonstrate the competitiveness of our method against the more costly RGB-D approaches.

cs.CV↗

Pro-KD: Progressive Distillation by Following the Footsteps of the Teacher

With ever growing scale of neural models, knowledge distillation (KD) attracts more attention as a prominent tool for neural model compression. However, there are counter intuitive observations in the literature showing some challenging limitations of KD. A case in point is that the best performing checkpoint of the teacher might not necessarily be the best teacher for training the student in KD. Therefore, one important question would be how to find the best checkpoint of the teacher for distillation? Searching through the checkpoints of the teacher would be a very tedious and computationally expensive process, which we refer to as the \textit{checkpoint-search problem}. Moreover, another observation is that larger teachers might not necessarily be better teachers in KD which is referred to as the \textit{capacity-gap} problem. To address these challenging problems, in this work, we introduce our progressive knowledge distillation (Pro-KD) technique which defines a smoother training path for the student by following the training footprints of the teacher instead of solely relying on distilling from a single mature fully-trained teacher. We demonstrate that our technique is quite effective in mitigating the capacity-gap problem and the checkpoint search problem. We evaluate our technique using a comprehensive set of experiments on different tasks such as image classification (CIFAR-10 and CIFAR-100), natural language understanding tasks of the GLUE benchmark, and question answering (SQuAD 1.1 and 2.0) using BERT-based models and consistently got superior results over state-of-the-art techniques.

cs.CL↗