arXiv subjects

Kanchan Sarkar

Publications and source records attributed to Kanchan Sarkar.

13 recordsLinked to original sources

arXiv2026

Linear Response Predicts Cusp-Pair Births in Networks with a Localized Cubic

Linear response is cheap to measure; the bistability boundaries it organizes are not. For a passive network with one localized cubic, the driving-point receptance $G$ fixes the period-one cusp set at fundamental-harmonic order: cusps lie on a fixed phase contour of $G$, a tangency of that contour under parameter variation creates a pair, and its curvature separates a gap opening from an isolated loop. For a two-mode absorber the linear prediction locates a benchmark birth coupling to $0.3\%$, and to $0.03\%$ once a third-harmonic correction of scale $|G(3Ω)/G(Ω)|$ is included.

nlin.CD↗

arXiv2026

Fifth-Harmonic Feedback Controls the Existence of a Third-Harmonic Zero

An exact zero of a generated harmonic needs two real cancellations; a damped oscillator's leading fundamental-source balance supplies one. Absorber loss leaves a phase deficit of fixed sign, amplitude independent at leading order, closed instead by the orbit's own $5Ω$ motion returning to $3Ω$ at weight $|Δ_1|^4$. Closure is therefore unreachable at weak amplitude: the channel supplying it ranks at $O(F^7)$, not at the leading $O(F^3)$. The fifth harmonic is $1.8\times10^{-3}$ of the relative-coordinate fundamental, yet a balance homotopy weighting it below $0.48$ annihilates the tracked pair in a saddle-node.

nlin.CD↗

arXiv2026

MLT-Dedup: Efficient Large-Scale Online Video Deduplication via Multi-Level Representations and Spatial-Temporal Matching

The explosive growth of user-generated video content on online platforms is accompanied by the emergence of numerous near-duplicate videos--videos that are identical or highly similar but differ by partial edits. These duplicates degrade user experience and increase storage and bandwidth costs, making large-scale video deduplication a critical task. Existing video deduplication frameworks face a fundamental challenge in retrieving sufficient high-quality candidates under a limited index budget, as well as trade-offs between efficiency and precision. To address these issues, we propose MLT-Dedup, an efficient large-scale online video deduplication framework with Multi-Level representations and spatial-Temporal matching. Our approach employs a Multi-Level Video Encoder (ML-VE) to extract both fine-grained frame-level and sparse clip-level embeddings: sparse embeddings support efficient candidate retrieval, while fine-grained embeddings are loaded for precise pairwise matching. During matching, we introduce DiF-SiM, a Differential Feature-enhanced Similarity Module capable of locating duplicated temporal segments and providing reliable similarity evidence to support policy-driven deduplication decisions. Extensive experiments on a real-world large-scale platform demonstrate that MLT-Dedup reduces online repetition rates by 91% at 90% precision. Furthermore, our sparse retrieval design achieves a 5x increase in indexing capacity, enabling broader candidate coverage in real-world deployment.

cs.CV↗

arXiv2026

MatchLM2Lite: A Scalable MLLM-to-Lite Framework for Reproduced Content Identification

Content moderation is critical for online video platforms to ensure content safety, protect creators, and sustain positive user experiences. Beyond filtering harmful content, platforms must guarantee content authenticity at scale so that users are exposed to diverse, original videos rather than low-value reproductions. We present MatchLM2Lite, a real-time, production-grade reproduced content identification (RCI) system that leverages the powerful understanding of a multimodal large language model (MLLM) distilled into a small and fast-inference model. Our system jointly models video, audio, and text signals, operating on pairs of videos to produce fine-grained reproduction scores. The system comprises two modules, MatchLM and MatchLite, and a two-stage training recipe. First, our high-capacity MLLM, MatchLM, serves as a teacher model to define the upper bound of RCI performance. Its capabilities are then distilled into a compact student model, MatchLite. This design allows MatchLite to deliver low-latency, high-throughput inference on video pairs while preserving much of MatchLM's accuracy, making it suitable for integration into real-time recommendation systems. MatchLM achieves an F1-score improvement of +8.57 compared to our previous production model. After knowledge distillation, MatchLite retains a +6.55 gain in F1-score while reducing computational cost by 35x. Deployed at scale, MatchLM2Lite enables efficient, pairwise multimodal RCI, stably serving online traffic at high queries per second (QPS) with an end-to-end latency below 30 seconds. This system has reduced the reproduced video view rate on our platform by 2.5% without degrading user engagement, demonstrating its effectiveness in a large-scale production environment.

cs.MM↗

arXiv2026

Dynamic Content Moderation in Livestreams: Combining Supervised Classification with MLLM-Boosted Similarity Matching

Content moderation remains a critical yet challenging task for large-scale user-generated video platforms, especially in livestreaming environments where moderation must be timely, multimodal, and robust to evolving forms of unwanted content. We present a hybrid moderation framework deployed at production scale that combines supervised classification for known violations with reference-based similarity matching for novel or subtle cases. This hybrid design enables robust detection of both explicit violations and novel edge cases that evade traditional classifiers. Multimodal inputs (text, audio, visual) are processed through both pipelines, with a multimodal large language model (MLLM) distilling knowledge into each to boost accuracy while keeping inference lightweight. In production, the classification pipeline achieves 67% recall at 80% precision, and the similarity pipeline achieves 76% recall at 80% precision. Large-scale A/B tests show a 6-8% reduction in user views of unwanted livestreams}. These results demonstrate a scalable and adaptable approach to multimodal content governance, capable of addressing both explicit violations and emerging adversarial behaviors.

cs.CV↗

arXiv2026

CAMEL: Confidence-Gated Reflection for Reward Modeling

Reward models play a fundamental role in aligning large language models with human preferences. Existing methods predominantly follow two paradigms: scalar discriminative preference models, which are efficient but lack interpretability, and generative judging models, which offer richer reasoning at the cost of higher computational overhead. We observe that the log-probability margin between verdict tokens strongly correlates with prediction correctness, providing a reliable proxy for instance difficulty without additional inference cost. Building on this insight, we propose CAMEL, a confidence-gated reflection framework that performs a lightweight single-token preference decision first and selectively invokes reflection only for low-confidence instances. To induce effective self-correction, we train the model via reinforcement learning with counterfactual prefix augmentation, which exposes the model to diverse initial verdicts and encourages genuine revision. Empirically, CAMEL achieves state-of-the-art performance on three widely used reward-model benchmarks with 82.9% average accuracy, surpassing the best prior model by 3.2% and outperforming 70B-parameter models using only 14B parameters, while establishing a strictly better accuracy-efficiency Pareto frontier.

cs.CL↗

arXiv2026

DISCOVER: A Physics-Informed, GPU-Accelerated Symbolic Regression Framework

Symbolic Regression (SR) enables the discovery of interpretable mathematical relationships from experimental and simulation data. These relationships are often coined descriptors which are defined as a fundamental materials property that is directly correlated to a desired or undesired functional property of the material. Although established approaches such as Sure Independence Screening and Sparsifying Operator (SISSO) have successfully identified low-dimensional descriptors within large feature spaces many existing SR tools integrate poorly with modern Python workflows, offer limited control over the symbolic search space, or struggle with the computational demands of large-scale studies. This paper introduces DISCOVER (Data-Informed Symbolic Combination of Operators for Variable Equation Regression), an open-source symbolic regression package developed to address these challenges through a modular, physics-motivated design. DISCOVER allows users to guide the symbolic search using domain knowledge, constrain the feature space explicitly, and take advantage of optional GPU acceleration to improve computational efficiency in data-intensive workflows, enabling reproducible and scalable SR workflows. The software is intended for applications in computational physics, computational chemistry, and materials science, where interpretability, physical consistency, and execution time are especially important, and it complements general-purpose SR frameworks by emphasizing the discovery of physically meaningful models.

physics.comp-ph↗

arXiv2025

FOCUS: Efficient Keyframe Selection for Long Video Understanding

Multimodal large language models (MLLMs) represent images and video frames as visual tokens. Scaling from single images to hour-long videos, however, inflates the token budget far beyond practical limits. Popular pipelines therefore either uniformly subsample or apply keyframe selection with retrieval-style scoring using smaller vision-language models. However, these keyframe selection methods still rely on pre-filtering before selection to reduce the inference cost and can miss the most informative moments. We propose FOCUS, Frame-Optimistic Confidence Upper-bound Selection, a training-free, model-agnostic keyframe selection module that selects query-relevant frames under a strict token budget. FOCUS formulates keyframe selection as a combinatorial pure-exploration (CPE) problem in multi-armed bandits: it treats short temporal clips as arms, and uses empirical means and Bernstein confidence radius to identify informative regions while preserving exploration of uncertain areas. The resulting two-stage exploration-exploitation procedure reduces from a sequential policy with theoretical guarantees, first identifying high-value temporal regions, then selecting top-scoring frames within each region. On two long-video question-answering benchmarks, FOCUS delivers substantial accuracy improvements while processing less than 2% of video frames. For videos longer than 20 minutes, it achieves an 11.9% gain in accuracy on LongVideoBench, demonstrating its effectiveness as a keyframe selection method and providing a simple and general solution for scalable long-video understanding with MLLMs. Code is available at https://github.com/NUS-HPC-AI-Lab/FOCUS.

cs.CV↗

arXiv2025

Info-Coevolution: An Efficient Framework for Data Model Coevolution

Machine learning relies heavily on data, yet the continuous growth of real-world data poses challenges for efficient dataset construction and training. A fundamental yet unsolved question is: given our current model and data, does a new data (sample/batch) need annotation/learning? Conventional approaches retain all available data, leading to non-optimal data and training efficiency. Active learning aims to reduce data redundancy by selecting a subset of samples to annotate, while it increases pipeline complexity and introduces bias. In this work, we propose Info-Coevolution, a novel framework that efficiently enables models and data to coevolve through online selective annotation with no bias. Leveraging task-specific models (and open-source models), it selectively annotates and integrates online and web data to improve datasets efficiently. For real-world datasets like ImageNet-1K, Info-Coevolution reduces annotation and training costs by 32\% without performance loss. It is able to automatically give the saving ratio without tuning the ratio. It can further reduce the annotation ratio to 50\% with semi-supervised learning. We also explore retrieval-based dataset enhancement using unlabeled open-source data. Code is available at https://github.com/NUS-HPC-AI-Lab/Info-Coevolution/.

cs.LG↗

arXiv2022

Orbital-ordered ferromagnetic insulating state in tensile-strained SrCoO$_{3}$ thin films

At ambient pressure, bulk SrCoO$_{3}$ is a ferromagnetic (FM) metal in cubic perovskite structure. By contrast, magnetic properties of epitaxial SrCoO$_{3}$ thin films, especially at high tensile strain ($\varepsilon \gtrsim 3$\%), remain unclear: Previous calculations had predicted antiferromagnetic (AFM) states more energetically favorable in this regime, but recent experiments indicated a FM insulating state. In this work, using first-principles calculations, we perform an extensive search for the structural, spin, magnetic, and orbital states of SrCoO$_{3}$ thin films. Our calculations indicate that at $0 < \varepsilon \lesssim 2.5$\%, SrCoO$_{3}$ favors a FM half-metallic state with intermediate-spin ($t_{2g}^{5}e_{g}^{1}$-like) Co exhibiting $d^{6}\underline L$ character. At $\varepsilon \gtrsim 2.5$\%, a FM insulating state with high-spin ($t_{2g}^{4}e_{g}^{2}$-like) Co dominates. This FM insulating state is achieved via complicated orbital ordering, cooperative Jahn--Teller distortion, and octahedral tilting about all three crystal axes.

cond-mat.mtrl-sci↗

arXiv2022

The postperovskite transition in Fe- and Al-bearing bridgmanite: effects on seismic observables

The primary phase of the Earth's lower mantle, (Al, Fe)-bearing bridgmanite, transitions to the postperovskite (PPv) phase at Earth's deep mantle conditions. Despite extensive experimental and ab initio investigations, there are still important aspects of this transformation that need clarification. Here, we address this transition in (Al3+, Fe3+)-, (Al3+)-, (Fe2+)-, and (Fe3+)-bearing bridgmanite using ab initio calculations and validate our results against experiments on similar compositions. Consistent with experiments, our results show that the onset transition pressure and the width of the two-phase region depend distinctly on the chemical composition: a) Fe3+-, Al3+-, or (Al3+, Fe3+)-alloying increases the transition pressure, while Fe2+-alloying has the opposite effect; b) in the absence of coexisting phases, the pressure-depth range of the Pv-PPv transition seems quite broad to cause a sharp D" discontinuity (< 30 km); c) the average Clapeyron slope of the two-phase regions are consistent with previous measurements, calculations in MgSiO3, and inferences from seismic data. In addition, d) we observe a softening of the bulk modulus in the two-phase region. The consistency between our results and experiments gives us the confidence to proceed and examine this transition in aggregates with different compositions computationally, which will be fundamental for resolving the most likely chemical composition of the D" region by analyses of tomographic images.

cond-mat.mtrl-sci↗

arXiv2016

Evolutionary optimization of PAW data-sets for accurate high pressure simulations

We examine the challenge of performing accurate electronic structure calculations at high pressures by comparing the results of all-electron full potential linearized augmented-plane-wave calculations with those of the projector augmented wave (PAW) method. In particular, we focus on developing an automated and consistent way of generating transferable PAW data-sets that can closely produce the all electron equation of state defined from zero to arbitrary high pressures. The technique we propose is an evolutionary search procedure that exploits the ATOMPAW code to generate atomic data-sets and the Quantum ESPRESSO software suite for total energy calculations. We demonstrate different aspects of its workability by optimizing PAW basis functions of some elements relatively abundant in planetary interiors. In addition, we introduce a new measure of atomic data-set goodness by considering their performance uniformity over an enlarged pressure range.

physics.comp-ph↗

arXiv2015

Pure and Hybrid Evolutionary Computing in Global Optimization of Chemical Structures: from Atoms and Molecules to Clusters and Crystals

The growth of evolutionary computing (EC) methods in the exploration of complex potential energy landscapes of atomic and molecular clusters, as well as crystals over the last decade or so is reviewed. The trend of growth indicates that pure as well as hybrid evolutionary computing techniques in conjunction of DFT has been emerging as a powerful tool, although work on molecular clusters has been rather limited so far. Some attempts to solve the atomic/molecular Schrodinger Equation (SE) directly by genetic algorithms (GA) are available in literature. At the Born-Oppenheimer level of approximation GA-density methods appear to be a viable tool which could be more extensively explored in the coming years, specially in the context of designing molecules and materials with targeted properties.

cond-mat.mtrl-sci↗