arXiv subjects

Md Adnan Arefeen

Publications and source records attributed to Md Adnan Arefeen.

13 recordsLinked to original sources

arXiv2026

ChitraMiti: Benchmarking Visual Grounding and Modality Reliance in Bengali Geometric Reasoning

Evaluation of vision-language models (VLMs) for multimodal mathematical reasoning remains limited for low-resource languages and for geometry problems that require reading a diagram and a question together. We introduce ChitraMiti-12.8k, a synthetic benchmark of 12,874 Bengali planar geometry problems paired with structured 15-attribute descriptions, and NCTB-500, a complementary set of 500 diagrams manually extracted from Bengali school textbooks. Using a three-phase protocol that separates diagram-only, diagram-plus-description, and description-only inputs, we show across five open-weight and closed-source VLMs that description-only performance is statistically indistinguishable from diagram-plus-description performance, establishing structured descriptions as a sufficient textual proxy for controlled evaluation. Despite this, models remain poor at cross-modal verification, frequently misled by a swapped spatial relation even when they answer the unmodified item correctly. We further evaluate supervised adaptation on ChitraMiti-12.8k, finding that fine-tuning improves performance on both ChitraMiti-1k and NCTB-500, although a substantial gap to the strongest zero-shot model remains. Together, ChitraMiti-12.8k, NCTB-500, and our evaluation protocol offer a standardized way to study Bengali multimodal geometry reasoning and, more broadly, whether VLMs actually check their text against what they see. Our dataset and code are publicly available on Hugging Face at https://huggingface.co/datasets/RaiyanKhaan/ChitraMiti.

cs.CV↗

arXiv2026

Do Small Models Use the Law You Give Them? Measuring Context Use on a Bilingual Bangladesh Legal Benchmark

Fine-tuning can improve legal question-answering accuracy without improving how models use law supplied in context. We study this distinction in bilingual Bangladeshi legal QA, where observed errors can arise from answer scoring, retrieval, or failure to use relevant law. We construct a hierarchy-preserving statutory corpus, 2,165 reviewed bilingual fine-tuning examples, and a 150-item supplied-law control. We evaluate six instruction-tuned models: Llama-3.2-1B, Llama-3.2-3B, Qwen3.5-0.8B, Qwen3.5-2B, Qwen3.5-4B, and Gemma-4-E2B, with three LoRA seeds per model. To separate effects, we combine constrained option-letter scoring, cyclic option rotation, and controlled removal of the governing provision. On 398 Bar Council outputs, an exact-line parser attributes an accuracy gain of 50.0\% to the Qwen3.5-2B seed-42 adapter, whereas option scoring yields only $3.0\%$. For Gemma-4-E2B, the two scoring methods favor different systems. When the governing provision is guaranteed to be present, five of six reference models improve by $14.7\%-19.3\%$ under the four-order criterion. Removing that provision reduces accuracy by $8.0\%-15.3\%$ for models and by $13.8\%-14.9\%$ points for their adapters. However, difference-in differences estimates show no increase in reliance on the governing provision after fine-tuning. Results show that legal adaptation claims require separating scorer, retriever, and model effects. Our Code and data are available at https://anonymous.4open.science/r/bangladesh-legal-qa-11E3

cs.CL↗

arXiv2026

Time Present and Time Past: Benchmarking Large Language Models on Temporally Evolving Document Understanding

Evolving documents, such as laws, tax codes, and software documentation, are amended, replaced, and sometimes reverted over time, so a question has different correct answers at different dates. In contrast to encyclopedic knowledge, where an old fact is simply overwritten, an amendment is itself an official text that states what it replaces and when it takes effect, and the earlier version stays correct for its validity period. The central challenge is therefore version resolution, that is, identifying the version in force on the queried date. Existing temporal QA datasets treat time only as an annotation, so version resolution stays untested. We present TIDE, an expert-verified benchmark of 3,050 QA pairs over 644 official customs instruments issued between 1969 and 2025 by the Government of Bangladesh, covering eight task types over deeply code-mixed documents that are heterogeneous in layout and dated in two calendars. In addition, we evaluate nine recent LLMs under a single protocol across parametric, gold-context, and retrieval access, scored by a three-judge LLM council with a hard date gate separating correct meaning from correct time. The best macro-averaged accuracy is only 68.5%. Resolving a version from an implicit date reaches 59.7%, and detecting that the supplied version does not govern the query reaches only 26.7%. Models are more likely to find correct versions than to reject incorrect ones, and they tend to follow a confident parametric answer over the supplied authoritative text. All code and data are available at https://github.com/icsetepa44/TIDE

cs.AI↗

arXiv2026

QuanTiMedAI: Quantum-Enhanced Time-Series Model guided by Agentic AI for Cardiac Arrest Mortality Prediction

Cardiac arrest remains one of the most lethal conditions encountered in intensive care units. Despite the growing availability of electronic health record data, existing mortality prediction studies in this population largely depend on static summaries derived from early admission. Such approaches ignore the temporal progression of physiological deterioration and recovery that unfolds throughout a patient's ICU stay. To address this limitation, we introduce QuanTiMedAI, a quantum-agentic framework developed for cardiac arrest mortality prediction using agentic AI guided quantum enhancement time series model. The proposed system combines an agentic large language model (LLM) for clinically informed feature discovery with a compact quantum recurrent network for temporality aware mortality prediction. Our findings demonstrate that agentic LLM-guided feature selection consistently outperforms conventional feature selection approaches, and the proposed quantum architecture achieves competitive predictive performance through nonlinear feature enhancement while keeping the number of parameters very low. Through extensive experimentation on a MIMIC-IV cohort of cardiac arrest patients, QuanTiMedAI's quantum-enhanced architecture attains an AUROC of 0.852 using only 605 parameters, an improvement of approximately 2.9\% over a current state-of-the-art baseline for this task. A structured ablation study systematically validates the contribution of each architectural design choice. These results show that quantum-enhanced sequential modeling can exceed classical recurrent networks while using substantially fewer parameters.

cs.AI↗

arXiv2026

Do Small Models Use the Law You Give Them? Context-Injected Fine-Tuning for Legal QA in Bangladesh

A small language model can receive the governing statutory provision and still answer incorrectly. We test whether fine-tuning on examples containing relevant law improves later use of retrieved law. We curate 2{,}165 bilingual QA records from six Bangladeshi acts and three schedules, then fine-tune Qwen3.5 at 0.8B, 2B, and 4B. Evaluation uses the 2022 and 2023 Bangladesh Bar Council exams in Bangla and machine-translated English, with no retrieval, BM25, or FAISS, scored by strict consistency over three seeded runs. At 0.8B, fine-tuning raises the 2022 English FAISS score from 2 to 34 of 100. Gains at 0.8B and 2B survive paired testing, but the 4B model has no detectable net gain: Bangla improves while several English conditions regress. Fine-tuning also reduces answers that drift from Bangla into mostly English from 44.0--53.2\% to 0.2--0.7\%, with adjusted $p<.001$ at every scale. Retrieval quality is therefore not the only bottleneck. Small bilingual legal models also differ in how they use supplied law and whether they answer in the requested language. The dataset is publicly available at https://huggingface.co/datasets/momahadi/bangladesh-legal-qa-dataset.

cs.CL↗

arXiv2026

Open-SAT: LLM-Guided Query Embedding Refinement for Open-Vocabulary Object Retrieval in Satellite Imagery

In satellite applications, user queries often take the form of open-ended natural language, extending beyond a fixed set of predefined categories. This open-vocabulary nature poses significant challenges for retrieving relevant image tiles, as the retrieval system must generalize to a wide range of unseen objects and concepts. While vision-language models (VLMs) such as CLIP are widely used for text-image retrieval, even fine-tuned variants often struggle to accurately align such queries with satellite imagery. To address this, we propose Open-SAT, a training-free query embedding refinement algorithm that operates at inference time to improve alignment between user queries and satellite image content. Open-SAT uses VLMs to compute embeddings for image tiles, which are stored in a vector database for efficient retrieval. At query time, it leverages Large Language Models (LLMs) to refine the text embeddings by incorporating contextual information about objects of interest and their surroundings. A threshold-free retrieval mechanism further enhances accuracy and efficiency. Experimental results in three public benchmarks demonstrate that Open-SAT improves the F1 score by up to 16.04%, while retrieving a comparable number of image tiles. These results demonstrate the effectiveness of Open-SAT in open-vocabulary satellite image retrieval, leveraging LLM guidance without the need for additional training or supervision.

cs.CV↗

arXiv2025

TrafficLens: Multi-Camera Traffic Video Analysis Using LLMs

Traffic cameras are essential in urban areas, playing a crucial role in intelligent transportation systems. Multiple cameras at intersections enhance law enforcement capabilities, traffic management, and pedestrian safety. However, efficiently managing and analyzing multi-camera feeds poses challenges due to the vast amount of data. Analyzing such huge video data requires advanced analytical tools. While Large Language Models (LLMs) like ChatGPT, equipped with retrieval-augmented generation (RAG) systems, excel in text-based tasks, integrating them into traffic video analysis demands converting video data into text using a Vision-Language Model (VLM), which is time-consuming and delays the timely utilization of traffic videos for generating insights and investigating incidents. To address these challenges, we propose TrafficLens, a tailored algorithm for multi-camera traffic intersections. TrafficLens employs a sequential approach, utilizing overlapping coverage areas of cameras. It iteratively applies VLMs with varying token limits, using previous outputs as prompts for subsequent cameras, enabling rapid generation of detailed textual descriptions while reducing processing time. Additionally, TrafficLens intelligently bypasses redundant VLM invocations through an object-level similarity detector. Experimental results with real-world datasets demonstrate that TrafficLens reduces video-to-text conversion time by up to $4\times$ while maintaining information accuracy.

cs.CV↗

arXiv2024

iRAG: Advancing RAG for Videos with an Incremental Approach

Retrieval-augmented generation (RAG) systems combine the strengths of language generation and information retrieval to power many real-world applications like chatbots. Use of RAG for understanding of videos is appealing but there are two critical limitations. One-time, upfront conversion of all content in large corpus of videos into text descriptions entails high processing times. Also, not all information in the rich video data is typically captured in the text descriptions. Since user queries are not known apriori, developing a system for video to text conversion and interactive querying of video data is challenging. To address these limitations, we propose an incremental RAG system called iRAG, which augments RAG with a novel incremental workflow to enable interactive querying of a large corpus of videos. Unlike traditional RAG, iRAG quickly indexes large repositories of videos, and in the incremental workflow, it uses the index to opportunistically extract more details from select portions of the videos to retrieve context relevant to an interactive user query. Such an incremental workflow avoids long video to text conversion times, and overcomes information loss issues due to conversion of video to text, by doing on-demand query-specific extraction of details in video data. This ensures high quality of responses to interactive user queries that are often not known apriori. To the best of our knowledge, iRAG is the first system to augment RAG with an incremental workflow to support efficient interactive querying of a large corpus of videos. Experimental results on real-world datasets demonstrate 23x to 25x faster video to text ingestion, while ensuring that latency and quality of responses to interactive user queries is comparable to responses from a traditional RAG where all video data is converted to text upfront before any user querying.

cs.CV↗

arXiv2023

LeanContext: Cost-Efficient Domain-Specific Question Answering Using LLMs

Question-answering (QA) is a significant application of Large Language Models (LLMs), shaping chatbot capabilities across healthcare, education, and customer service. However, widespread LLM integration presents a challenge for small businesses due to the high expenses of LLM API usage. Costs rise rapidly when domain-specific data (context) is used alongside queries for accurate domain-specific LLM responses. One option is to summarize the context by using LLMs and reduce the context. However, this can also filter out useful information that is necessary to answer some domain-specific queries. In this paper, we shift from human-oriented summarizers to AI model-friendly summaries. Our approach, LeanContext, efficiently extracts $k$ key sentences from the context that are closely aligned with the query. The choice of $k$ is neither static nor random; we introduce a reinforcement learning technique that dynamically determines $k$ based on the query and context. The rest of the less important sentences are reduced using a free open source text reduction method. We evaluate LeanContext against several recent query-aware and query-unaware context reduction approaches on prominent datasets (arxiv papers and BBC news articles). Despite cost reductions of $37.29\%$ to $67.81\%$, LeanContext's ROUGE-1 score decreases only by $1.41\%$ to $2.65\%$ compared to a baseline that retains the entire context (no summarization). Additionally, if free pretrained LLM-based summarizers are used to reduce context (into human consumable summaries), LeanContext can further modify the reduced context to enhance the accuracy (ROUGE-1 score) by $13.22\%$ to $24.61\%$.

cs.CL↗

arXiv2023

MetaMorphosis: Task-oriented Privacy Cognizant Feature Generation for Multi-task Learning

With the growth of computer vision applications, deep learning, and edge computing contribute to ensuring practical collaborative intelligence (CI) by distributing the workload among edge devices and the cloud. However, running separate single-task models on edge devices is inefficient regarding the required computational resource and time. In this context, multi-task learning allows leveraging a single deep learning model for performing multiple tasks, such as semantic segmentation and depth estimation on incoming video frames. This single processing pipeline generates common deep features that are shared among multi-task modules. However, in a collaborative intelligence scenario, generating common deep features has two major issues. First, the deep features may inadvertently contain input information exposed to the downstream modules (violating input privacy). Second, the generated universal features expose a piece of collective information than what is intended for a certain task, in which features for one task can be utilized to perform another task (violating task privacy). This paper proposes a novel deep learning-based privacy-cognizant feature generation process called MetaMorphosis that limits inference capability to specific tasks at hand. To achieve this, we propose a channel squeeze-excitation based feature metamorphosis module, Cross-SEC, to achieve distinct attention of all tasks and a de-correlation loss function with differential-privacy to train a deep learning model that produces distinct privacy-aware features as an output for the respective tasks. With extensive experimentation on four datasets consisting of diverse images related to scene understanding and facial attributes, we show that MetaMorphosis outperforms recent adversarial learning and universal feature generation methods by guaranteeing privacy requirements in an efficient way for image and video analytics.

cs.CV↗

arXiv2022

FrameHopper: Selective Processing of Video Frames in Detection-driven Real-Time Video Analytics

Detection-driven real-time video analytics require continuous detection of objects contained in the video frames using deep learning models like YOLOV3, EfficientDet. However, running these detectors on each and every frame in resource-constrained edge devices is computationally intensive. By taking the temporal correlation between consecutive video frames into account, we note that detection outputs tend to be overlapping in successive frames. Elimination of similar consecutive frames will lead to a negligible drop in performance while offering significant performance benefits by reducing overall computation and communication costs. The key technical questions are, therefore, (a) how to identify which frames to be processed by the object detector, and (b) how many successive frames can be skipped (called skip-length) once a frame is selected to be processed. The overall goal of the process is to keep the error due to skipping frames as small as possible. We introduce a novel error vs processing rate optimization problem with respect to the object detection task that balances between the error rate and the fraction of frames filtering. Subsequently, we propose an off-line Reinforcement Learning (RL)-based algorithm to determine these skip-lengths as a state-action policy of the RL agent from a recorded video and then deploy the agent online for live video streams. To this end, we develop FrameHopper, an edge-cloud collaborative video analytics framework, that runs a lightweight trained RL agent on the camera and passes filtered frames to the server where the object detection model runs for a set of applications. We have tested our approach on a number of live videos captured from real-life scenarios and show that FrameHopper processes only a handful of frames but produces detection results closer to the oracle solution and outperforms recent state-of-the-art solutions in most cases.

cs.CV↗

arXiv2021

EARLIN: Early Out-of-Distribution Detection for Resource-efficient Collaborative Inference

Collaborative inference enables resource-constrained edge devices to make inferences by uploading inputs (e.g., images) to a server (i.e., cloud) where the heavy deep learning models run. While this setup works cost-effectively for successful inferences, it severely underperforms when the model faces input samples on which the model was not trained (known as Out-of-Distribution (OOD) samples). If the edge devices could, at least, detect that an input sample is an OOD, that could potentially save communication and computation resources by not uploading those inputs to the server for inference workload. In this paper, we propose a novel lightweight OOD detection approach that mines important features from the shallow layers of a pretrained CNN model and detects an input sample as ID (In-Distribution) or OOD based on a distance function defined on the reduced feature space. Our technique (a) works on pretrained models without any retraining of those models, and (b) does not expose itself to any OOD dataset (all detection parameters are obtained from the ID training dataset). To this end, we develop EARLIN (EARLy OOD detection for Collaborative INference) that takes a pretrained model and partitions the model at the OOD detection layer and deploys the considerably small OOD part on an edge device and the rest on the cloud. By experimenting using real datasets and a prototype implementation, we show that our technique achieves better results than other approaches in terms of overall accuracy and cost when tested against popular OOD datasets on top of popular deep learning models pretrained on benchmark datasets.

cs.CV↗

arXiv2021

A Lightweight ReLU-Based Feature Fusion for Aerial Scene Classification

In this paper, we propose a transfer-learning based model construction technique for the aerial scene classification problem. The core of our technique is a layer selection strategy, named ReLU-Based Feature Fusion (RBFF), that extracts feature maps from a pretrained CNN-based single-object image classification model, namely MobileNetV2, and constructs a model for the aerial scene classification task. RBFF stacks features extracted from the batch normalization layer of a few selected blocks of MobileNetV2, where the candidate blocks are selected based on the characteristics of the ReLU activation layers present in those blocks. The feature vector is then compressed into a low-dimensional feature space using dimension reduction algorithms on which we train a low-cost SVM classifier for the classification of the aerial images. We validate our choice of selected features based on the significance of the extracted features with respect to our classification pipeline. RBFF remarkably does not involve any training of the base CNN model except for a few parameters for the classifier, which makes the technique very cost-effective for practical deployments. The constructed model despite being lightweight outperforms several recently proposed models in terms of accuracy for a number of aerial scene datasets.

eess.IV↗