Yogesh Kulkarni

Yogesh Kulkarni, Pooyan Fazli

Recent advances in video-large language models (Video-LLMs) have led to significant progress in video understanding. Current preference optimization methods often rely on proprietary APIs or human-annotated captions to generate preference data (i.e., pairs of model outputs ranked by quality or alignment with human judgment), which is then used to train models for video-language alignment. This approach is both costly and labor-intensive. To address this limitation, we introduce VideoSAVi (Self-Aligned Video Language Model), a self-training pipeline that enables Video-LLMs to learn from video content without external supervision. Our approach includes a self-critiquing mechanism that identifies reasoning errors in the model's initial responses and generates improved alternatives, creating preference pairs directly from video content. VideoSAVi then applies Direct Preference Optimization (DPO) to iteratively train the model using the preference data, thus enhancing its temporal and spatial reasoning for video understanding. Experiments show that VideoSAVi delivers significant improvements across multiple benchmarks, including a +4.2 percentage point gain on MVBench, +3.9 on PerceptionTest, and +6.8 on the challenging EgoSchema dataset compared to baseline models. Our model-agnostic approach is computationally efficient, requiring only 32 frames, offering a promising direction for self-aligned video understanding without reliance on external models or annotations.

Yogesh Kulkarni, Pooyan Fazli

Multimodal reasoning over long-horizon video is challenging due to the need for precise spatiotemporal fusion and alignment across modalities. While recent methods such as Group Relative Policy Optimization (GRPO) have shown promise in this domain, they suffer from three key limitations: (1) data inefficiency from their on-policy design, (2) a vanishing advantage problem, where identical or near-identical rewards within a group eliminate the learning signal by producing zero-valued advantages, and (3) uniform credit assignment that fails to emphasize critical reasoning steps. We introduce AVATAR (Audio-Video Agent for Alignment and Reasoning), a framework that addresses these limitations through two core components: (1) an off-policy training architecture that improves sample efficiency and resolves vanishing advantages by reusing past experiences with greater reward diversity, and (2) Temporal Advantage Shaping (TAS), a novel credit assignment strategy that upweights key reasoning phases during learning. AVATAR achieves strong performance across various benchmarks, outperforming the Qwen2.5-Omni baseline by +5.4on MMVU, +4.9 on OmniBench, and +4.5 on Video-Holmes, while demonstrating over 35% higher sample efficiency.

Yogesh Kulkarni, Pooyan Fazli

Video-language models (Video-LLMs) excel at understanding video content but struggle with spatial relationships, temporal ordering, and cross-frame continuity. To address these limitations, we introduce VideoPASTA (Preference Alignment with Spatio-Temporal-Cross Frame Adversaries), a framework that enhances Video-LLMs through targeted preference optimization. VideoPASTA trains models to distinguish accurate video representations from carefully crafted adversarial examples that deliberately violate spatial, temporal, or cross-frame relationships. With only 7,020 preference pairs and Direct Preference Optimization, VideoPASTA enables models to learn robust representations that capture fine-grained spatial details and long-range temporal dynamics. Experiments demonstrate that VideoPASTA is model agnostic and significantly improves performance, for example, achieving gains of up to +3.8 percentage points on LongVideoBench, +4.1 on VideoMME, and +4.0 on MVBench, when applied to various state-of-the-art Video-LLMs. These results demonstrate that targeted alignment, rather than massive pretraining or architectural modifications, effectively addresses core video-language challenges. Notably, VideoPASTA achieves these improvements without any human annotation or captioning, relying solely on 32-frame sampling. This efficiency makes our approach a scalable plug-and-play solution that seamlessly integrates with existing models while preserving their original capabilities.

Yogesh Kulkarni, Pooyan Fazli

Reasoning about intentions and actions from a first-person (egocentric) perspective remains a fundamental challenge for multimodal large language models (MLLMs). Unlike third-person (exocentric) videos that capture scenes from an outside observer, egocentric videos reflect the actor's continuously changing viewpoint, introducing partial observability, limited field of view, and self-referenced motion. We introduce $\textbf{EgoVITA}$, a reinforcement learning framework that enables MLLMs to reason through structured planning and verification. Built on Group Relative Policy Optimization (GRPO), EgoVITA alternates between two stages: (1) an $\textbf{egocentric planning phase}$, where the model reasons from a first-person viewpoint to predict a step-by-step plan of future actions, and (2) an $\textbf{exocentric verification phase}$, where it switches to a third-person perspective to check the visual and logical consistency of that plan. Through GRPO, the model learns to make plans that are causally predictive of upcoming visual observations, leading to more coherent and visually grounded reasoning. EgoVITA achieves significant gains on egocentric reasoning tasks, outperforming the baseline Qwen2.5-VL-7B by $\mathbf{+7.7}$ on EgoBlind and $\mathbf{+4.4}$ on EgoOrient, while maintaining strong generalization on exocentric video tasks.

Chaoyu Li, Yogesh Kulkarni, Pooyan Fazli

The computational cost of training multimodal large language models (MLLMs) rapidly increases with the number of tokens involved. Existing efficiency methods primarily target inference and rely on token reduction or merging, offering limited benefit during training. In this paper, we propose ReGATE (Reference$-$Guided Adaptive Token Elision), an adaptive token pruning method for accelerating MLLM training. Specifically, ReGATE adopts a teacher-student framework in which the MLLM being trained serves as the student, and a frozen reference large language model (LLM) acts as the teacher. The teacher computes per-token reference losses, which are combined with an exponential moving average (EMA) of the student's own difficulty scores. This adaptive difficulty-based scoring enables the selective processing of crucial tokens while bypassing less informative ones in the forward pass, significantly reducing computational overhead. Experiments demonstrate that ReGATE, when applied to VideoLLaMA2, matches the peak accuracy of standard training on MVBench up to 2$\times$ faster, using only 35% of the tokens. With additional training, it even surpasses the baseline on several multimodal benchmarks, all while reducing the total token count by over 41%. Code and models will be released soon.

Yogesh Kulkarni, Sayf Hussain Z, Krithi Ramamritham et al.

Artificial intelligence-based techniques applied to the electricity consumption data generated from the smart grid prove to be an effective solution in reducing Non Technical Loses (NTLs), thereby ensures safety, reliability, and security of the smart energy systems. However, imbalanced data, consecutive missing values, large training times, and complex architectures hinder the real time application of electricity theft detection models. In this paper, we present EnsembleNTLDetect, a robust and scalable electricity theft detection framework that employs a set of efficient data pre-processing techniques and machine learning models to accurately detect electricity theft by analysing consumers' electricity consumption patterns. This framework utilises an enhanced Dynamic Time Warping Based Imputation (eDTWBI) algorithm to impute missing values in the time series data and leverages the Near-miss undersampling technique to generate balanced data. Further, stacked autoencoder is introduced for dimensionality reduction and to improve training efficiency. A Conditional Generative Adversarial Network (CTGAN) is used to augment the dataset to ensure robust training and a soft voting ensemble classifier is designed to detect the consumers with aberrant consumption patterns. Furthermore, experiments were conducted on the real-time electricity consumption data provided by the State Grid Corporation of China (SGCC) to validate the reliability and efficiency of EnsembleNTLDetect over the state-of-the-art electricity theft detection models in terms of various quality metrics.

Yogesh Kulkarni, Krisha Bhambani

With the Rise of Adversarial Machine Learning and increasingly robust adversarial attacks, the security of applications utilizing the power of Machine Learning has been questioned. Over the past few years, applications of Deep Learning using Deep Neural Networks(DNN) in several fields including Medical Diagnosis, Security Systems, Virtual Assistants, etc. have become extremely commonplace, and hence become more exposed and susceptible to attack. In this paper, we present a novel study analyzing the weaknesses in the security of deep learning systems. We propose 'Kryptonite', an adversarial attack on images. We explicitly extract the Region of Interest (RoI) for the images and use it to add imperceptible adversarial perturbations to images to fool the DNN. We test our attack on several DNN's and compare our results with state of the art adversarial attacks like Fast Gradient Sign Method (FGSM), DeepFool (DF), Momentum Iterative Fast Gradient Sign Method (MIFGSM), and Projected Gradient Descent (PGD). The results obtained by us cause a maximum drop in network accuracy while yielding minimum possible perturbation and in considerably less amount of time per sample. We thoroughly evaluate our attack against three adversarial defence techniques and the promising results showcase the efficacy of our attack.