Multimodal Large Language Models (MLLMs) have significantly advanced AI-assisted medical diagnosis, but they often generate factually inconsistent responses that deviate from established medical knowledge. Retrieval-Augmented Generation (RAG) enhances factual accuracy by integrating external sources, but it presents two key challenges. First, insufficient retrieval can miss critical information, whereas excessive retrieval can introduce irrelevant or misleading content, disrupting model output. Second, even when the model initially provides correct answers, over-reliance on retrieved data can lead to factual errors. To address these issues, we introduce the Multimodal Intelligent Retrieval and Augmentation (MIRA) framework, designed to optimize factual accuracy in MLLM. MIRA consists of two key components: (1) a calibrated Rethinking and Rearrangement module that dynamically adjusts the number of retrieved contexts to manage factual risk, and (2) A medical RAG framework integrating image embeddings and a medical knowledge base with a query-rewrite module for efficient multimodal reasoning. This enables the model to effectively integrate both its inherent knowledge and external references. Our evaluation of publicly available medical VQA and report generation benchmarks demonstrates that MIRA substantially enhances factual accuracy and overall performance, achieving new state-of-the-art results.

We focus on source-free domain adaptive object detection (SF-DAOD) problem when source data is unavailable during adaptation and the model must adapt to unlabeled target domain. Majority of approaches for the problem employ a self-supervised approach using a student-teacher (ST) framework where pseudo-labels are generated via a sourcepretrained model for further fine-tuning. We observe that the performance of a student model often degrades drastically, due to collapse of teacher model primarily caused by high noise in pseudo-labels, resulting from domain bias, discrepancies, and a significant domain shift across domains. To obtain reliable pseudo-labels, we propose a Target-based Iterative Query-Token Adversarial Network (TITAN) which separates the target images into two subsets that are similar to the source (easy) and those that are dissimilar (hard). We propose a strategy to estimate variance to partition the target domain. This approach leverages the insight that higher detection variances correspond to higher recall and greater similarity to the source domain. Also, we incorporate querytoken based adversarial modules into a student-teacher baseline framework to reduce the domain gaps between two feature representations. Experiments conducted on four natural imaging datasets and two challenging medical datasets have substantiated the superior performance of TITAN compared to existing state-of-the-art (SOTA) methodologies. We report an mAP improvement of +22.7, +22.2, +21.1 and +3.7 percent over the current SOTA on C2F, C2B, S2C, and K2C benchmarks respectively. .

We focus on the Source Free Object Detection (SFOD) problem, when source data is unavailable during adaptation, and the model must adapt to the unlabeled target domain. In medical imaging, several approaches have leveraged a semi-supervised student-teacher architecture to bridge domain discrepancy. Context imbalance in labeled training data and significant domain shifts between domains can lead to biased teacher models that produce inaccurate pseudolabels, degrading the student model's performance and causing a mode collapse. Class imbalance, particularly when one class significantly outnumbers another, leads to contextual bias. To tackle the problem of context bias and the significant performance drop of the student model in the SFOD setting, we introduce Grounded Teacher (GT) as a standard framework. In this study, we model contextual relationships using a dedicated relational context module and leverage it to mitigate inherent biases in the model. This approach enables us to apply augmentations to closely related classes, across and within domains, enhancing the performance of underrepresented classes while keeping the effect on dominant classes minimal. We further improve the quality of predictions by implementing an expert foundational branch to supervise the student model. We validate the effectiveness of our approach in mitigating context bias under the SFOD setting through experiments on three medical datasets supported by comprehensive ablation studies. All relevant resources, including preprocessed data, trained model weights, and code, are publicly available.

Large Language Models (LLMs) have transformed the natural language processing landscape and brought to life diverse applications. Pretraining on vast web-scale data has laid the foundation for these models, yet the research community is now increasingly shifting focus toward post-training techniques to achieve further breakthroughs. While pretraining provides a broad linguistic foundation, post-training methods enable LLMs to refine their knowledge, improve reasoning, enhance factual accuracy, and align more effectively with user intents and ethical considerations. Fine-tuning, reinforcement learning, and test-time scaling have emerged as critical strategies for optimizing LLMs performance, ensuring robustness, and improving adaptability across various real-world tasks. This survey provides a systematic exploration of post-training methodologies, analyzing their role in refining LLMs beyond pretraining, addressing key challenges such as catastrophic forgetting, reward hacking, and inference-time trade-offs. We highlight emerging directions in model alignment, scalable adaptation, and inference-time reasoning, and outline future research directions. We also provide a public repository to continually track developments in this fast-evolving field.

One of the severe and pressing problems which humanity is facing in the 21st Century is climate change. It is adversely affecting the entire globe as is evident with the rising ocean levels, accelerated melting of glaciers, frequent storms, and many more. The need of the hour is to limit climate change by reducing greenhouse gas emissions. Machine learning algorithms have proven to be extremely effective in numerous scenarios and the experts are working to extend their uses to limit the effects of climate change. The domain is vast but in the current literature, the works in this domain are very limited. In this paper, we have tried to work in this particular direction by applying deep learning methodologies to analyze the effect of different climate factors on overall climate change. We proposed a novel climate change parameter model to predict the correlation between different climate parameters. We evaluated our scheme by validating the model on the data specific to the Jammu and Kashmir region and the model loss was found to be under 2%. In addition, based on our model we proposed some reactive strategies that can be implemented to keep the climate change effects under control..

Cataract surgery is the most common surgical procedure globally, with a disproportionately higher burden in developing countries. While automated surgical video analysis has been explored in general surgery, its application to ophthalmic procedures remains limited. Existing works primarily focus on Phaco cataract surgery, an expensive technique not accessible in regions where cataract treatment is most needed. In contrast, Manual Small-Incision Cataract Surgery (MSICS) is the preferred low-cost, faster alternative in high-volume settings and for challenging cases. However, no dataset exists for MSICS. To address this gap, we introduce Sankara-MSICS, the first comprehensive dataset containing 53 surgical videos annotated for 18 surgical phases and 3,527 frames with 13 surgical tools at the pixel level. We benchmark this dataset on state-of-the-art models and present ToolSeg, a novel framework that enhances tool segmentation by introducing a phase-conditional decoder and a simple yet effective semi-supervised setup leveraging pseudo-labels from foundation models. Our approach significantly improves segmentation performance, achieving a 23.77% to 38.10% increase in mean Dice scores, with a notable boost for tools that are less prevalent and small. Furthermore, we demonstrate that ToolSeg generalizes to other surgical settings, showcasing its effectiveness on the CaDIS dataset.

One of the biggest issues facing humanity in the twenty-first century is climate change, as shown by the increasing sea levels, melting glaciers, and frequent storms. Accurate temperature forecasting is crucial for understanding and mitigating its impacts. Cutting-edge data-driven models for temperature forecasting typically employ recurrent neural networks(CNNs), with certain models integrating attention mechanisms. However RNNs sequential processing limits parallelization, especially for longer sequences. In order to do this, we provide a brand-new method for temperature prediction that is based on the FocalNet Transformer architecture. By functioning in a multi-tensor format, the suggested Focal-modulation Attention Encoder (FATE) framework leverages the spatial and temporal nuances of meteorological data characteristics by integrating tensorized modulation. Comparative assessments against existing transformer encoder architectures, 3D CNNs, LSTM, and ConvLSTM demonstrate our model’s superior ability to capture nuanced patterns inherent in the data, particularly in the context of temperature prediction. We also introduce a new labeled dataset, Climate change Parameter dataset (CCPD), which encompasses 40 years of data from J&K region on seven key parameters that influence climate change, supporting further research in this area. Experiments on two real-world benchmark temperature datasets from weather stations in the USA, Canada, and Europe demonstrate accuracy improvements of 12%, 23%, and 28% respectively, compared to existing SOTA models. In addition, we achieved state-of-the-art results on our CCPD dataset with a 24% improvement. To understand FATE, we introduce two modulation scores from the tensorial modulation process. These scores clarify our model’s decision making and key climate change parameters. For reproducible research, we will release the source code, pre-trained FATE model, and CCPD dataset.

In clinical applications, X-Ray technology plays a crucial role in noninvasive examinations like mammography, providing essential anatomical information about patients. However, the inherent radiation risk associated with X-Ray procedures raises significant concerns. X-Ray reconstruction is crucial in medical imaging for creating detailed visual representations of internal structures, and facilitating diagnosis and treatment without invasive procedures. Recent advancements in deep learning (DL) have shown promise in X-Ray reconstruction. Nevertheless, conventional DL methods often necessitate the centralized aggregation of substantial large datasets for training, following specific scanning protocols. This requirement results in notable domain shifts and privacy issues. To address these challenges, we introduce the Hierarchical Framework based Federated Learning method (HF-Fed) for customized X-Ray Imaging. HF-Fed addresses the challenges in X-Ray imaging optimization by decomposing the problem into two components: local data adaptation and holistic X-Ray Imaging. It employs a hospital-specific hierarchical framework and a shared common imaging network called Network of Networks (NoN) for these tasks. The emphasis of the NoN is on acquiring stable features from a variety of data distributions. A hierarchical hypernetwork extracts domain-specific hyperparameters, conditioning the NoN for customized X-Ray reconstruction. Experimental results demonstrate HF-Fed’s competitive performance, offering a promising solution for enhancing X-Ray imaging without the need for data sharing. This study significantly contributes to the evolving body of literature on the potential advantages of federated learning in the healthcare sector. It offers valuable insights for policymakers and healthcare providers holistically.

Human pose estimation is the process of continuously monitoring a person's action and movement to track and monitor the activity of a person or an object. Human pose estimation is usually done by capturing the key points which describe the pose of a person. A guiding practicing framework that enables people to learn and exercise activities like yoga, fittness, dancing, etc., might be built using human posture recognition remotely and accurately without the help of a personal trainer. This work has proposed a framework to detect and recognize various yoga and exercise poses to help the individual practice the same correctly. A popular Blaze-pose model extracts key points from the student end and compares the same with the instructor pose. The extracted key points are fed to the Human Pose Juxtaposition model (HPJT) to compare the student pose with the instructor. The model will assess the correctness of the pose by comparing the extracted key points and give feedback to students if any corrections need to be made. The proposed model is trained with 40+ yoga and exercise poses, and evaluated the model's performance with the mAP, and the model achieved an accuracy of 99.04%. The results proved that any person could use the proposed framework in real-time to practice exercise, yoga, dance, etc. At their respective location without the help of a physical instructor with precision and accuracy, leading to a healthy life.

In this paper, we present a Climate Change Parameter Dataset (CCPD) intending to achieve state-of-the-art results in parameters which effect climate change, including forest cover, water bodies, agriculture and vegetation, population, temperature, construction, and air index. The dataset can be used by the research community to validate the claims made in relation to the climate change. Research community has been deeply involved in extending the use case of machine learning algorithms to the effects of climate change. However, the non-availability of sufficient data related to climate change parameters has limited the research in this domain. By presenting this dataset, we want to facilitate the researchers. In this dataset, we provide a large variety of statistical and satellite data acquired by various image processing techniques and on-ground data collection. The data is collected in abundance for a specific region, and then various machine learning techniques are used to extract the useful data related to each parameter separately. We call this amalgam of processed data as CCPD dataset. CCPD dataset contains over 6000 data points for all seven parameters and covers the data from 1960 onwards. We hope this dataset will aid the research community in tackling climate change with the help of AI.

The area of Computer Vision has gone through exponential growth and advancement over the past decade. It is mainly due to the introduction of effective deep-learning methodologies and the availability of massive data. This has resulted in the incorporation of intelligent computer vision schemes to automate the different number of tasks. In this paper, we have worked on similar lines. We have proposed an integrated system for the development of robotic arms, considering the current situation in fruit identification, classification, counting, and generating their masks through semantic segmentation. The current method of manually doing these processes is time-consuming and is not feasible for large fields. Due to this, multiple works have been proposed to automate harvesting tasks to minimize the overall overhead. However, there is a lack of an integrated system that can automate all these processes together. As a result, we are proposing one such approach based on different machine learning techniques. For each process, we propose to use the most effective learning technique with computer vision capability. Thus, proposing an integrated intelligent end-to-end computer vision-based system to detect, classify, count, and identify the apples. In this system, we modified the YOLOv3 algorithm to detect and count the apples effectively. The proposed scheme works even under variable lighting conditions. The system was trained and tested using a standard benchmark i.e., MinneApple. Experimental results show an average accuracy of 91%..