Abstract

Verifying scientific claims presents a significantly greater challenge than verifying political or news-related claims. Unlike the relatively broad audience for political claims, the users of scientific claim verification systems can vary widely, ranging from researchers testing specific hypotheses to everyday users seeking information on a medication. Additionally, the evidence for scientific claims is often highly complex, involving technical terminology and intricate domain-specific concepts that require specialized models for accurate verification. Despite considerable interest from the research community, there is a noticeable lack of large-scale scientific claim verification datasets to benchmark and train effective models. To bridge this gap, we introduce two large-scale datasets, SciClaimHunt and SciClaimHunt_Num, derived from scientific research papers. We propose several baseline models tailored for scientific claim verification to assess the effectiveness of these datasets. Additionally, we evaluate models trained on SciClaimHunt and SciClaimHunt_Num against existing scientific claim verification datasets to gauge their quality and reliability. Furthermore, we conduct human evaluations of the claims in proposed datasets and perform error analysis to assess the effectiveness of the proposed baseline models. Our findings indicate that SciClaimHunt and SciClaimHunt_Num serve as highly reliable resources for training models in scientific claim verification

Abstract

We introduce RoadSocial, a large-scale, diverse VideoQA dataset tailored for generic road event understanding from social media narratives. Unlike existing datasets limited by regional bias, viewpoint bias and expert-driven annotations, RoadSocial captures the global complexity of road events with varied geographies, camera viewpoints (CCTV, handheld, drones) and rich social discourse. Our scalable semi-automatic annotation framework leverages Text LLMs and Video LLMs to generate comprehensive question-answer pairs across 12 challenging QA tasks, pushing the boundaries of road event understanding. RoadSocial is derived from social media videos spanning 14M frames and 414K social comments, resulting in a dataset with 13.2K videos, 674 tags and 260K high-quality QA pairs. We evaluate 18 Video LLMs (open-source and proprietary, driving-specific and general-purpose) on our road event understanding benchmark. We also demonstrate RoadSocial's utility in improving road event understanding capabilities of general-purpose Video LLMs.

Abstract

Deep neural networks have become invaluable intellectual property in machine learning. To deter model theft, Watermarking has emerged as a prominent defense by embedding hidden “trigger sets” that aid in ownership verification. However, current watermarking solutions primarily address scenarios where adversaries steal the entire model. In this paper, we reveal a critical gap: partial model extraction, where only a subset of classes is stolen, substantially degrading the watermark’s reliability. We introduce two attacks, Partial Model Extraction and Partial Knowledge Distillation, which reduce watermark accuracy by up to 80% while retaining strong performance on the stolen classes. Through extensive experiments on CIFAR10 and CIFAR100 against two state-of-the-art watermarking schemes, we demonstrate the need for more robust watermarking strategies that resist partial-class theft.

Abstract

Background: Sensory Processing Sensitivity (SPS), a temperament trait reflecting deeper cognitive processing and heightened emotional/physiological reactivity, is often linked to emotional dysregulation and vulnerability to stress-related problems. SPS includes three subscales: Ease of Excitation (EOE) reflects emotional sensitivity to internal (e.g., thoughts, bodily states) and external (e.g., social demands) stimuli; Low Sensory Threshold (LST) reflects sensitivity to intense sensory input; and Aesthetic Sensitivity (AES) denotes appreciation for subtleties (Smolewska et al., 2006). Music can sometimes contribute to maladaptive outcomes like rumination, avoidance coping, and amplification of negative affect (Alluri et al., 2022). Highly sensitive persons, especially with elevated EOE, may be more prone to such patterns due to their emotional reactivity. However, SPS’s interactive role in maladaptive music use remains underexplored. Aim: This study investigates the relationship between SPS and maladaptive music use, hypothesizing that individuals with high EOE, particularly those sensitive to external demands (External EOE), are more prone to unhealthy music engagement. Network analysis was employed to reveal how sensitivity traits, distress, and empathy interact in shaping music use patterns. Methods: A total of 680 Indian participants(mean age=20.69, s.d.=1.9; 180 females) completed a battery of self-report measures including Highly Sensitive Persons Scale (HSPS) for SPS traits, the Kessler Psychological Distress Scale (K10) (Andrews & Slade, 2001), Interpersonal Reactivity Index (IRI) for empathy traits (Keaton, 2017), and Healthy-Unhealthy Music Scale (Saarikallio et al., 2015) assessing adaptive and maladaptive music use. The IRI included four subscales: Empathic Concern (EC), Perspective Taking (PT), Personal Distress (PD), and Fantasy Scale (FS), capturing both cognitive and affective empathy. Using exploratory factor analysis we split EOE into Internal and External factors. Regularized Partial Correlation Networks (RPCNs) and Bayesian Networks were constructed to model interactions and infer relationships among variables. Results: Unhealthy music use correlated with External EOE(r=0.25) while Healthy music use correlated with AES(r=0.32) among SPS traits. RPCNs identified External EOE as a central node in the network(strength=0.69; betweenness=116), bridging SPS traits with distress-related personal feelings(PD from IRI), psychological symptoms (K10), and Unhealthy music use. Bayesian Networks confirmed K10 as a direct predictor of Unhealthy use(B=0.50), with PD predicting K10(B=0.31) and External EOE predicting PD(B=0.50). Mediation analysis revealed that K10 significantly mediated the pathway from External EOE to Unhealthy music use, accounting for 55% of the total effect (all p<0.01), suggesting distress amplifies the impact of emotional sensitivity on maladaptive behaviors. Discussion: While music can support emotion regulation, our results suggest that highly sensitive persons experiencing distress may turn to music as a maladaptive coping strategy to escape overwhelming external demands. In India, where population density and sensory overstimulation are high, such patterns may be intensified. These findings add to the emerging literature on trait-environment interactions in music behavior and suggest that understanding EOE may inform future interventions or wellness strategies tailored to individual sensitivity profiles.

Abstract

Sensory processing sensitivity (SPS), a trait marked by heightened reactivity to stimuli, is linked to emotional dysregulation and stress-related problems. Its subscales include Ease of Excitation that reflects emotional sensitivity to internal and external demands, Low Sensory Threshold reflects sensory overload susceptibility, and Aesthetic Sensitivity denotes appreciation for subtleties. While music can sometimes amplify maladaptive outcomes (rumination, avoidance), SPS’s role in such behaviours, especially in non-Western contexts, remains underexplored. This study examines how SPS drives maladaptive music listening in 673 Indian adults. Network analysis and structural equation modelling revealed sensitivity to external demands elevated psychological distress which in turn predicted maladaptive music use. Reactivity to internal demands may reduce maladaptive music use but exacerbated it when mediated by external demands, reflecting preference for avoidant coping in Indian context. Findings emphasize SPS’s role in maladaptive behaviours, demonstrating how sensory reactivity interacts with traits to shape distress-driven music use as an emotional regulation mechanism.

Abstract

Machine Learning Enabled Systems (MLS) are becoming integral to real-world applications, but ensuring their sustainable performance over time remains a significant challenge. These systems operate in dynamic environments and face runtime uncertainties like data drift and model degradation, which affect the sustainability of MLS across multiple dimensions: technical, economical, environmental, and social. While Machine Learning Operations (MLOps) addresses the technical dimension by streamlining the ML model lifecycle, it overlooks other dimensions. Furthermore, some traditional practices, such as frequent retraining, incur substantial energy and computational overhead, thus amplifying sustainability concerns. To address them, we introduce HarmonE, an architectural approach that enables self-adaptive capabilities in MLOps pipelines using the MAPE-K loop. HarmonE allows system architects to define explicit sustainability goals and adaptation thresholds at design time, and performs runtime monitoring of key metrics, such as prediction accuracy, energy consumption, and data distribution shifts, to trigger appropriate adaptation strategies. We validate our approach using a Digital Twin (DT) of an Intelligent Transportation System (ITS), focusing on traffic flow prediction as our primary use case. The DT employs time series ML models to simulate real-time traffic and assess various flow scenarios. Our results show that HarmonE adapts effectively to evolving conditions while maintaining accuracy and meeting sustainability goals.

Abstract

The robustness of precast concrete structures is primarily governed by connections between the precast structural components. Joint connections deliver a vital role in load transfer, restraining movement, and providing stability at local and global level. This calls for appropriate connection mechanism for precast elements providing integrity under different loading conditions and ease of construction. The connection details in literature or available in practice have been rarely evaluated. This paper discusses the numerical modelling of precast reinforced concrete structural wall-column with different joint connection systems and evaluates various seismic design parameters, simulated through in-plane lateral loading in a finite element software. The connection systems evaluated are reinforcement bar connection, headed bar connections, U-bar connection and steel wire loop connection, which are subsequently compared with a cast-in-situ reinforced concrete wall-column model having similar material and geometrical properties. The seismic behaviour of connections have been studied with regards to the damage pattern, stresses, lateral load capacity, base shear coefficient, stiffness, deformation characteristics, strength factors, and damage index. Among all the connection systems, U-bar connection and loop connection using steel wire ropes showed stresses within the yield limit, while other connection systems reached their yield capacity, indicating damage. Apparently, steel wire loop connection is found to be effective in transfer of lateral load between the precast reinforced concrete elements, as observed from the numerical analysis; whereas other connection systems yielded prior to the design load. However, all connections except the reinforcement bar connection surpassed design load prior to the peak limit.

Abstract

Model Merging (MM) has emerged as a promising approach to combine multiple fine-tuned models into a single model that maintains performance across tasks. However, its impact on (transferable) adversarial examples remains unexplored. Specifically, we investigate the affect of MM on the transferability of adversarial examples---a black-box, adversarial attack where adversarial examples generated for surrogate model successfully mislead target model. Through experiments on image classification models, we discover that, similar to recent findings of MM on backdoor attacks, where MM tends to sanitize the backdoor present in individual fine-tuned models, MM does not boost adversarial transferability. Our analysis reveals MM has no affect, and under certain conditions, decreased vulnerability to transferable adversarial examples. Consequently, our finding also support the notion of MM giving a "free lunch'' of adversarial robustness and provides important insights for designing more secure systems that employ MM.

Abstract

According to the cooperative principle (Grice 1975), utterances in a conversation adhere to the four maxims of conversational implicature – quantity, relation, quality and manner. An implicature arises when a maxim is flouted or contradicted, and the speaker is forced to infer a meaning. Consider the example: Speaker A: का िदलीप को परीकाएं आसान लग रही है? Kya Dilip ko parikshayein aasan lag rahi hain? (Does Dilip find the exam easy?) Speaker B: आज-कल वो सो नही पा रहा | [Aaj-kal wo so nahi pa raha.] (He hasn’t been able to sleep lately.) Speaker B flouts the maxim of relevance by not answering the question with a yes or no. Instead, his reply that Dilip has trouble sleeping implies that he doesn’t find the exams to be easy. While computational approaches have made some progress in English implicature benchmarks - GRICE (Zheng, 2021), IMPRESS (Jeretic 2020) and Conversational Implicature benchmark (CIB) (George and Mamidi, 2020), there are no benchmarks for Hindi. Hence, we propose a Hindi implicature benchmark- Indirect Questions Hi Implicature Benchmark - Toledo-Ronen (2020) found that translation- based methods lose nuances in argument evaluation. XCOPA benchmark (Ponti 2020) & Huang (2024) demonstrated that translated data hurts model performance for structurally divergent non- English languages. Hindi is a free word order, morphologically rich language. It is structurally different from English hence, translating English implicature benchmarks may not be unsuitable. Wang (2024) present indirect answers to yes-no questions in dialogues benchmark. Similar to their approach, we plan to collect Hindi interviews and extract indirect questions along with their answers involving implicature. We will annotate 5000 such questions. We will maintain a constraint that previous dialogues will be added to the question as a background to simplify the inference over a single dialogue turn. Evaluation will rely on ROUGE scores to measure correctness, as we leave exploration of suitable metrics for future work. For model evaluation, we plan to leverage a) English LLMs (meta-llama/Llama-3.3-70B (Grattafiori 2024) and the current state-of-the-art reported by Sravanthi (2024), FlanT5-XXL (Chung 2022) finetuned on Hindi translated GRICE (Zheng 2021), IMPRESS (Jeretic 2020) and CIB (George and Mamidi 2020) and b) Hindi LLMs (Llama-3-Nanda-10B-Chat (Choudhury 2024) and Airavata (Gala 2024)).

Abstract

In recent years, widespread internet adoption and the growth in user base of various social media platforms have led to an increase in the prolifera tion of extreme speech online. While traditional language models have demon strated proficiency in distinguishing between neutral text and non-neutral text (i.e. extreme speech), categorizing the diverse types of extreme speech presents significant challenges [1][2]. The task of extreme speech classification is par ticularly nuanced, as it requires a deep understanding of socio-cultural contexts to accurately interpret the intent of the language used by the speaker. Even hu man annotators often disagree on the appropriate classification of such content, emphasizing the complex and subjective nature of this task [3]. The use of hu man moderators also presents a scaling issue, necessitating the need for auto mated systems for extreme speech classification. The recent launch of ChatGPT has drawn global attention to the potential applications of Large Language Models (LLMs) across a diverse variety of tasks. Trained on vast and diverse corpora, and demonstrating the ability to effectively capture and encode contex tual information, LLMs emerge as highly promising tools for tackling this spe cific task of extreme speech classification. In this paper, we leverage the Indian subset of the extreme speech dataset from [3] to develop an effective classifica tion framework using LLMs. We evaluate open-source Llama models against closed-source OpenAI models, finding that while pre-trained LLMs show mod erate efficacy, fine-tuning with domain-specific data significantly enhances per formance, highlighting their adaptability to linguistic and contextual nuances. Although GPT-based models outperform Llama models in zero-shot settings, the performance gap disappears after fine-tuning.