arXiv subjects

Manikandan Ravikiran

Publications and source records attributed to Manikandan Ravikiran.

16 recordsLinked to original sources

arXiv2026

When Models Defer to Wrong Answers: A Robustness Audit of Source-Attributed Cues in Multiple-Choice QA

Language models often receive a question together with a claim about what another source answered. We audit whether such claims destabilize answers in multiple-choice question answering. For each item, we hold one wrong option fixed across misleading conditions and vary the cue template attached to it. We introduce \emph{neutral-conditioned misleading cue adoption rate} (NC-MCAR), which measures switches to that option only on valid cued trials where the same model first selected the gold answer under a neutral prompt. This is a measure of answer instability, not proof that the model knew the answer or that all deference is irrational. We evaluate four instruction-following models on MMLU-Pro and IndicMMLU-Pro in English, Hindi, Bengali, Tamil, and Telugu. Across 220{,}000 outputs, the expert template yields 41.1\% aggregate NC-MCAR, compared with 12.5\% for the majority template. These two conditions use the same wrong option and final instruction. Filler accuracy remains well above expert-wrong accuracy, while correct-cue prompts have high valid-response accuracy. The audit documents answer instability relevant to grounding under the tested forced-choice prompts: a bare, unverified source claim can outweigh an answer that was previously consistent with the task evidence.

cs.CL↗

arXiv2026

The Audit Decides the Verdict: Instrument Effects Rival Demographic Bias in LLM Decision Audits

Whether a language model looks demographically biased can depend on how the audit asks its question. A charitable-aid benchmark reports that the same models favor minority applicants when rating requests one at a time and penalize some when ranking side by side. We test whether that reversal generalizes to hiring, lending, and medical triage: 40,726 requests to five models, applications differing only in the applicant's name, and a primary test fixed before collection. It does not. None of 36 planned contrasts survives correction. The rating advantage keeps its sign at roughly half the published size, and a precision extension bounds any hiring ranking penalty below the published effect, though the lending and triage ranking floors sit above that margin, so the exclusion is conclusive for hiring ranking and for rating in all three domains only. Planted disparities tracking their injected sizes and a directional replication on the original aid materials bound these nulls. The audit is livelier than the demographics: models recognize transparent audits nearly always, tie every identical-content comparison whether the varying detail is race or a hobby, and reward first-listed candidates as much as any demographic effect we measure. Audit verdicts reflect audit construction more than demographic bias.

cs.CL↗

arXiv2026

NVExplain: Explaining Time Series Forecasting with Latent Trajectory Analysis and Structure-Preserving Surrogates

Time series forecasting models are widely used in high-stakes settings, yet their predictions remain difficult to interpret because existing post-hoc methods often ignore temporal dependence and fail to provide horizon-specific explanations. We propose a model-agnostic explainability framework that explains forecasting predictions by attributing each forecast horizon to temporally relevant historical lags. The framework models forecasting as a latent trajectory and introduces semantic flow to quantify how information evolves across time in the model's internal representations. By aggregating semantic flow, it constructs a lag-horizon attribution matrix that captures horizon-resolved temporal influence. To improve explainability, we further generate structure-preserving perturbations and fit sparse local surrogate models, producing human-readable and temporally coherent explanations. We evaluate the method using faithfulness and stability diagnostics across multiple benchmark datasets. Results show that the semantic-flow variant achieves competitive or superior faithfulness compared to standard post-hoc baselines, while being substantially more computationally efficient. Stability analysis further demonstrates that the explanations are robust and identifies regimes where interpretation should be applied with caution.

cs.LG↗

arXiv2026

Refusal-Gated Decoding: Preserving Refusal Behavior Under High-Temperature Sampling

High-temperature sampling is one of the primary mechanisms for increasing diversity in LLMs. Recent advances in truncation-based sampling techniques have helped mitigate drawbacks of high-temperature sampling such as neural text degeneration, thereby enabling greater diversity in LLM outputs without sacrificing coherence. However, increasing the entropy of the token probability distribution via high temperatures has also been shown to weaken model guardrails by reducing the model's refusal response in the presence of harmful prompts. Despite the potential benefits of high-temperature sampling and the importance of maintaining model safety, there is a lack of existing solutions for maintaining the refusal behavior of LLMs under a higher entropy regime. To address this gap, we systematically study how temperature influences refusal behavior in LLMs and propose an efficient sequential decoding approach which preserves a model's greedy decoding refusal response at high temperatures while incurring minimal additional latency. Through extensive experiments, we show that our approach preserves 91-99% of the greedy decoding refusal behavior across three benchmark datasets without compromising the model's high-temperature response for safe prompts. Our work demonstrates how refusal behavior can be maintained in an efficient manner for applications which require high-temperature sampling.

cs.AI↗

arXiv2026

Riemannian-Manifold Steering: Geometry-Aware Generative Autoencoders for Label-Free Steering

Steering a language model - intervening on its internal activations to change downstream behaviour - has recently expanded beyond linear interpolation to nonlinear methods such as angular and kernelized steering, which define intervention transformations without learning an explicit geometry over paths in activation space. Freshly introduced geometry-aware manifold methods do learn such a geometry, but require labelled class centroids together with prescribed cyclic or sequential structure. These assumptions restrict where manifold steering can be applied, since existing constructions require labelled centroids and compatible boundary conditions. We recast manifold steering more broadly as \textbf{Riemannian geodesic computation} on activation space, recovering linear and labelled-spline steering as geodesics under particular choices of metric. A principled metric within this framework is the output-space Hellinger distance pulled back to activations; we approximate this with a learned encoder trained on output distances over a small concept-token schema - no per-prompt labels, no topology prior, and no per-task curve fitting. Empirically, the method reliably drives the model onto the target class across all tasks in a standard four-task language-model arithmetic benchmark, while following more behaviourally natural trajectories than baselines on smaller output spaces. We thereby provide a unified Riemannian framework for manifold steering together with a schema-supervised, label-free instantiation that operates without labelled centroids or prescribed boundary conditions.

cs.LG↗

arXiv2026

INDIC DIALECT: A Multi Task Benchmark to Evaluate and Translate in Indian Language Dialects

Recent NLP advances focus primarily on standardized languages, leaving most low-resource dialects under-served especially in Indian scenarios. In India, the issue is particularly important: despite Hindi being the third most spoken language globally (over 600 million speakers), its numerous dialects remain underrepresented. The situation is similar for Odia, which has around 45 million speakers. While some datasets exist which contain standard Hindi and Odia languages, their regional dialects have almost no web presence. We introduce INDIC-DIALECT, a human-curated parallel corpus of 13k sentence pairs spanning 11 dialects and 2 languages: Hindi and Odia. Using this corpus, we construct a multi-task benchmark with three tasks: dialect classification, multiple-choice question (MCQ) answering, and machine translation (MT). Our experiments show that LLMs like GPT-4o and Gemini 2.5 perform poorly on the classification task. While fine-tuned transformer based models pretrained on Indian languages substantially improve performance e.g., improving F1 from 19.6\% to 89.8\% on dialect classification. For dialect to language translation, we find that hybrid AI model achieves highest BLEU score of 61.32 compared to the baseline score of 23.36. Interestingly, due to complexity in generating dialect sentences, we observe that for language to dialect translation the ``rule-based followed by AI" approach achieves best BLEU score of 48.44 compared to the baseline score of 27.59. INDIC-DIALECT thus is a new benchmark for dialect-aware Indic NLP, and we plan to release it as open source to support further work on low-resource Indian dialects.

cs.CL↗

arXiv2025

Towards Transparent AI Grading: Semantic Entropy as a Signal for Human-AI Disagreement

Automated grading systems can efficiently score short-answer responses, yet they often fail to indicate when a grading decision is uncertain or potentially contentious. We introduce semantic entropy, a measure of variability across multiple GPT-4-generated explanations for the same student response, as a proxy for human grader disagreement. By clustering rationales via entailment-based similarity and computing entropy over these clusters, we quantify the diversity of justifications without relying on final output scores. We address three research questions: (1) Does semantic entropy align with human grader disagreement? (2) Does it generalize across academic subjects? (3) Is it sensitive to structural task features such as source dependency? Experiments on the ASAP-SAS dataset show that semantic entropy correlates with rater disagreement, varies meaningfully across subjects, and increases in tasks requiring interpretive reasoning. Our findings position semantic entropy as an interpretable uncertainty signal that supports more transparent and trustworthy AI-assisted grading workflows.

cs.AI↗

arXiv2022

Findings of the Shared Task on Offensive Span Identification from Code-Mixed Tamil-English Comments

Offensive content moderation is vital in social media platforms to support healthy online discussions. However, their prevalence in codemixed Dravidian languages is limited to classifying whole comments without identifying part of it contributing to offensiveness. Such limitation is primarily due to the lack of annotated data for offensive spans. Accordingly, in this shared task, we provide Tamil-English code-mixed social comments with offensive spans. This paper outlines the dataset so released, methods, and results of the submitted systems

cs.CL↗

arXiv2022

Zero-shot Code-Mixed Offensive Span Identification through Rationale Extraction

This paper investigates the effectiveness of sentence-level transformers for zero-shot offensive span identification on a code-mixed Tamil dataset. More specifically, we evaluate rationale extraction methods of Local Interpretable Model Agnostic Explanations (LIME) \cite{DBLP:conf/kdd/Ribeiro0G16} and Integrated Gradients (IG) \cite{DBLP:conf/icml/SundararajanTY17} for adapting transformer based offensive language classification models for zero-shot offensive span identification. To this end, we find that LIME and IG show baseline $F_{1}$ of 26.35\% and 44.83\%, respectively. Besides, we study the effect of data set size and training process on the overall accuracy of span identification. As a result, we find both LIME and IG to show significant improvement with Masked Data Augmentation and Multilabel Training, with $F_{1}$ of 50.23\% and 47.38\% respectively. \textit{Disclaimer : This paper contains examples that may be considered profane, vulgar, or offensive. The examples do not represent the views of the authors or their employers/graduate schools towards any person(s), group(s), practice(s), or entity/entities. Instead they are used to emphasize only the linguistic research challenges.}

cs.CL↗

arXiv2022

Overlapping Word Removal is All You Need: Revisiting Data Imbalance in Hope Speech Detection

Hope Speech Detection, a task of recognizing positive expressions, has made significant strides recently. However, much of the current works focus on model development without considering the issue of inherent imbalance in the data. Our work revisits this issue in hope-speech detection by introducing focal loss, data augmentation, and pre-processing strategies. Accordingly, we find that introducing focal loss as part of Multilingual-BERT's (M-BERT) training process mitigates the effect of class imbalance and improves overall F1-Macro by 0.11. At the same time, contextual and back-translation-based word augmentation with M-BERT improves results by 0.10 over baseline despite imbalance. Finally, we show that overlapping word removal based on pre-processing, though simple, improves F1-Macro by 0.28. In due process, we present detailed studies depicting various behaviors of each of these strategies and summarize key findings from our empirical results for those interested in getting the most out of M-BERT for hope speech detection under real-world conditions of data imbalance.

cs.CL↗

arXiv2020

Investigating the Effect of Intraclass Variability in Temporal Ensembling

Temporal Ensembling is a semi-supervised approach that allows training deep neural network models with a small number of labeled images. In this paper, we present our preliminary study on the effect of intraclass variability on temporal ensembling, with a focus on seed size and seed type, respectively. Through our experiments we find that (a) there is a significant drop in accuracy with datasets that offer high intraclass variability, (b) more seed images offer consistently higher accuracy across the datasets, and (c) seed type indeed has an impact on the overall efficiency, where it produces a spectrum of accuracy both lower and higher. Additionally, based on our experiments, we also find KMNIST to be a competitive baseline for temporal ensembling.

cs.CV↗

arXiv2020

Hitachi at SemEval-2020 Task 12: Offensive Language Identification with Noisy Labels using Statistical Sampling and Post-Processing

In this paper, we present our participation in SemEval-2020 Task-12 Subtask-A (English Language) which focuses on offensive language identification from noisy labels. To this end, we developed a hybrid system with the BERT classifier trained with tweets selected using Statistical Sampling Algorithm (SA) and Post-Processed (PP) using an offensive wordlist. Our developed system achieved 34 th position with Macro-averaged F1-score (Macro-F1) of 0.90913 over both offensive and non-offensive classes. We further show comprehensive results and error analysis to assist future research in offensive language identification with noisy labels.

cs.CL↗

arXiv2020

Finding Black Cat in a Coal Cellar -- Keyphrase Extraction & Keyphrase-Rubric Relationship Classification from Complex Assignments

Diversity in content and open-ended questions are inherent in complex assignments across online graduate programs. The natural scale of these programs poses a variety of challenges across both peer and expert feedback including rogue reviews. While the identification of relevant content and associating it to predefined rubrics would simplify and improve the grading process, the research to date is still in a nascent stage. As such in this paper we aim to quantify the effectiveness of supervised and unsupervised approaches for the task for keyphrase extraction and generic/specific keyphrase-rubric relationship extraction. Through this study, we find that (i) unsupervised MultiPartiteRank produces the best result for keyphrase extraction (ii) supervised SVM classifier with BERT features that offer the best performance for both generic and specific keyphrase-rubric relationship classification. We finally present a comprehensive analysis and derive useful observations for those interested in these tasks for the future. The source code is released in \url{https://github.com/manikandan-ravikiran/cs6460-proj}.

cs.CL↗

arXiv2020

Key Phrase Classification in Complex Assignments

Complex assignments typically consist of open-ended questions with large and diverse content in the context of both classroom and online graduate programs. With the sheer scale of these programs comes a variety of problems in peer and expert feedback, including rogue reviews. As such with the hope of identifying important contents needed for the review, in this work we present a very first work on key phrase classification with a detailed empirical study on traditional and most recent language modeling approaches. From this study, we find that the task of classification of key phrases is ambiguous at a human level producing Cohen's kappa of 0.77 on a new data set. Both pretrained language models and simple TFIDF SVM classifiers produce similar results with a former producing average of 0.6 F1 higher than the latter. We finally derive practical advice from our extensive empirical and model interpretability results for those interested in key phrase classification from educational reports in the future.

cs.CL↗

arXiv2020

What's happened in MOOC Posts Analysis, Knowledge Tracing and Peer Feedbacks? A Review

Learning Management Systems (LMS) and Educational Data Mining (EDM) are two important parts of online educational environment with the former being a centralised web-based information systems where the learning content is managed and learning activities are organised (Stone and Zheng,2014) and latter focusing on using data mining techniques for the analysis of data so generated. As part of this work, we present a literature review of three major tasks of EDM (See section 2), by identifying shortcomings and existing open problems, and a Blumenfield chart (See section 3). The consolidated set of papers and resources so used are released in https://github.com/manikandan-ravikiran/cs6460-Survey. The coverage statistics and review matrix of the survey are as shown in Figure 1 & Table 1 respectively. Acronym expansions are added in the Appendix Section 4.1.

cs.HC↗

arXiv2020

Systematic Review of Approaches to Improve Peer Assessment at Scale

Peer Assessment is a task of analysis and commenting on student's writing by peers, is core of all educational components both in campus and in MOOC's. However, with the sheer scale of MOOC's & its inherent personalised open ended learning, automatic grading and tools assisting grading at scale is highly important. Previously we presented survey on tasks of post classification, knowledge tracing and ended with brief review on Peer Assessment (PA), with some initial problems. In this review we shall continue review on PA from perspective of improving the review process itself. As such rest of this review focus on three facets of PA namely Auto grading and Peer Assessment Tools (we shall look only on how peer reviews/auto-grading is carried), strategies to handle Rogue Reviews, Peer Review Improvement using Natural Language Processing. The consolidated set of papers and resources so used are released in https://github.com/manikandan-ravikiran/cs6460-Survey-2.

cs.CY↗