arXiv subjects

Anu Agarwal

Publications and source records attributed to Anu Agarwal.

5 recordsLinked to original sources

arXiv2025

SIMU: Selective Influence Machine Unlearning

The undesired memorization of sensitive information by Large Language Models (LLMs) has emphasized the need for safety mechanisms that can regulate model behavior. This has led to the development of machine unlearning techniques that enable models to precisely forget sensitive and unwanted information. For machine unlearning, first-order and second-order optimizer-based methods have shown significant progress in enabling LLMs to forget targeted information. However, in doing so, these approaches often compromise the model's original capabilities, resulting in unlearned models that struggle to retain their prior knowledge and overall utility. To address this, we propose Selective Influence Machine Unlearning (SIMU), a two-step framework that enhances second-order optimizer-based unlearning by selectively updating only the critical neurons responsible for encoding the forget-set. By constraining updates to these targeted neurons, SIMU achieves comparable unlearning efficacy while substantially outperforming current methods in retaining the model's original knowledge.

cs.LG↗

arXiv2020

Cluster Based Deep Contextual Reinforcement Learning for top-k Recommendations

Rapid advancements in the E-commerce sector over the last few decades have led to an imminent need for personalised, efficient and dynamic recommendation systems. To sufficiently cater to this need, we propose a novel method for generating top-k recommendations by creating an ensemble of clustering with reinforcement learning. We have incorporated DB Scan clustering to tackle vast item space, hence in-creasing the efficiency multi-fold. Moreover, by using deep contextual reinforcement learning, our proposed work leverages the user features to its full potential. With partial updates and batch updates, the model learns user patterns continuously. The Duelling Bandit based exploration provides robust exploration as compared to the state-of-art strategies due to its adaptive nature. Detailed experiments conducted on a public dataset verify our claims about the efficiency of our technique as com-pared to existing techniques.

cs.IR↗

arXiv2020

Integrated Optofluidic Sensor for Coagulation Risk Monitoring in COVID-19 Patients at Point-of-Care

While the pathophysiology underlying the COVID-19 infection remains incompletely understood, there is growing evidence to indicate that it is closely correlated to hypercoagulation among severely ill patients. Doctors may choose for use anti-coagulation doses to treat the patients at intensive care units. A rapid, easy, and low-cost solution to monitor the coagulation status at the point-of-care may help with treatment by enabling the administration of controlled doses of medication to patients and to understand the disease's underlying pathophysiology. Thromboelastography, the clinical standard is accurate; it suffers from limited portability and low sensitivity when miniaturized to handheld form factor. In the article, we summarize research helping to advance towards an integrated optofluidic device combining microfluidics and photonic sensor technology. Microfluidics are used to perform blood pre-processing, and a photonic sensor measures blood coagulation status in real-time readout on the device itself. These techniques make it portable and scalable, potentially serving as technology foundation for the development of a disposable sensor for point-of-care diagnostics in COVID-19 patients and coagulopathy in general.

physics.med-ph↗

arXiv2020

Detection of Bio-aerosols and COVID-19 Equivalent Particles Via On-chip Mid Infrared Photonic Spectroscopy

We propose an on-chip mid-infrared (MIR) photonic spectroscopy platform for aerosol characterization to obtain highly discriminatory information on the chemistry of aerosol particles. Sensing of aerosols is crucial for various environmental, climactic, warfare threat detection, and pulmonary healthcare applications. Further, there are a number of unintended situations for potential exposure to bioaerosols such as viruses, bacteria, and fungi. For instance, the current pandemic scenario of COVID-19 occurring across the world. Currently, chemical characterization of aerosols is performed using FTIR spectroscopy yielding chemical fingerprinting because most of the vibrational and rotational transitions of chemical molecules fall in the MIR range; and Raman spectroscopy. Both techniques use free space bench-top geometries. Here, we propose miniaturized on-chip MIR photonics-based aerosol spectroscopy consisting of a broadband spiral-waveguide sensor that significantly enhances particle-light interaction to improve sensitivity. The spiral waveguides are made of a chalcogenide glass material (Ge23Sb7S70) which shows a broad transparency over IR range. We demonstrate the sensing of N-methyl aniline-based aerosol particles with the device. We anticipate that the sensor will readily complement existing photonic resonator-based particle sizing and counting techniques to develop a unified framework for on-chip integrated photonic aerosol spectroscopy.

physics.app-ph↗

arXiv2018

Chemical Characterization of Aerosol Particles Using On-chip Photonic Cavity Enhanced Spectroscopy

We demonstrate the chemical characterization of aerosol particles with on-chip spectroscopy using a photonic cavity enhanced silicon nitride (Si3N4) racetrack resonator-based sensor. The sensor operates over a broad and continuous wavelength range, showing cavity enhanced sensitivity at specific resonant wavelengths. Analysis of the relative change in the quality factor of the cavity resonances successfully yields the absorption spectrum of the aerosol particles deposited on the resonators. Detection of N-methyl aniline-based aerosol detection in the Near InfraRed (NIR) range of 1500 nm to 1600 nm is demonstrated. Our aerosol sensor spectral data compares favorably with that from a commercial spectrometer, indicating good accuracy. The small size of the device is advantageous in remote, environmental, medical and body-wearable sensing applications.

physics.app-ph↗