arXiv subjects

Balachandra Devarangadi Sunil

Publications and source records attributed to Balachandra Devarangadi Sunil.

3 recordsLinked to original sources

arXiv2026

Task-Aware LoRA Adapter Composition via Similarity Retrieval in Vector Databases

Parameter efficient fine tuning methods like LoRA have enabled task specific adaptation of large language models, but efficiently composing multiple specialized adapters for unseen tasks remains challenging. We present a novel framework for dynamic LoRA adapter composition that leverages similarity retrieval in vector databases to enable zero-shot generalization across diverse NLP tasks. Our approach constructs a task-aware vector database by embedding training examples from 22 datasets spanning commonsense reasoning, question answering, natural language inference, and sentiment analysis. At inference time, we retrieve the most similar training examples, compute task similarity distributions via nucleus sampling, and dynamically merge relevant LoRA adapters using retrieval weighted fusion strategies. We evaluated four merging methods Linear, Concatenation, TIES, and Magnitude Prune demonstrating that our dataset centric retrieval approach often matches or exceeds the performance of individually fine-tuned task-specific adapters. Notably, Linear merging achieves 70.95% on PIQA and 77.62% on RTE, substantially outperforming single-task baselines (46% and 52%, respectively). Our framework requires no additional retriever training, operates with frozen embeddings, and enables efficient, interpretable adapter composition. These results suggest that retrieval based dynamic merging offers a promising direction for scalable, parameter-efficient multitask learning without requiring full model retraining for each new task.

cs.CL↗

arXiv2026

Memory Poisoning Attack and Defense on Memory Based LLM-Agents

Large language model agents equipped with persistent memory are vulnerable to memory poisoning attacks, where adversaries inject malicious instructions through query only interactions that corrupt the agents long term memory and influence future responses. Recent work demonstrated that the MINJA (Memory Injection Attack) achieves over 95 % injection success rate and 70 % attack success rate under idealized conditions. However, the robustness of these attacks in realistic deployments and effective defensive mechanisms remain understudied. This work addresses these gaps through systematic empirical evaluation of memory poisoning attacks and defenses in Electronic Health Record (EHR) agents. We investigate attack robustness by varying three critical dimensions: initial memory state, number of indication prompts, and retrieval parameters. Our experiments on GPT-4o-mini, Gemini-2.0-Flash and Llama-3.1-8B-Instruct models using MIMIC-III clinical data reveal that realistic conditions with pre-existing legitimate memories dramatically reduce attack effectiveness. We then propose and evaluate two novel defense mechanisms: (1) Input/Output Moderation using composite trust scoring across multiple orthogonal signals, and (2) Memory Sanitization with trust-aware retrieval employing temporal decay and pattern-based filtering. Our defense evaluation reveals that effective memory sanitization requires careful trust threshold calibration to prevent both overly conservative rejection (blocking all entries) and insufficient filtering (missing subtle attacks), establishing important baselines for future adaptive defense mechanisms. These findings provide crucial insights for securing memory-augmented LLM agents in production environments.

cs.CR↗

arXiv2025

Density Estimation and Crowd Counting

This study enhances a crowd density estimation algorithm originally designed for image-based analysis by adapting it for video-based scenarios. The proposed method integrates a denoising probabilistic model that utilizes diffusion processes to generate high-quality crowd density maps. To improve accuracy, narrow Gaussian kernels are employed, and multiple density map outputs are generated. A regression branch is incorporated into the model for precise feature extraction, while a consolidation mechanism combines these maps based on similarity scores to produce a robust final result. An event-driven sampling technique, utilizing the Farneback optical flow algorithm, is introduced to selectively capture frames showing significant crowd movements, reducing computational load and storage by focusing on critical crowd dynamics. Through qualitative and quantitative evaluations, including overlay plots and Mean Absolute Error (MAE), the model demonstrates its ability to effectively capture crowd dynamics in both dense and sparse settings. The efficiency of the sampling method is further assessed, showcasing its capability to decrease frame counts while maintaining essential crowd events. By addressing the temporal challenges unique to video analysis, this work offers a scalable and efficient framework for real-time crowd monitoring in applications such as public safety, disaster response, and event management.

cs.CV↗