Browsing by Author "Sinha, Yash"

Now showing 1 - 18 of 18

Addressing Challenges in Browser Based P2P Content Sharing Framework Using WebRTC
(IEEE, 2016) Haribabu, K; Sinha, Yash
Most of the content sharing applications use the client/server model in which all of group managements are done by the server and this sometimes becomes a communication bottleneck. Installing specialized software for different purposes such as file sharing, video conferencing etc., becomes a barrier for the user. Recent technologies like NodeJs and Socket.io have fostered new ideas the ways web browsers can be used. Moreover, the emerging standards of WebRTC open up new paradigm of direct communication channel between web browsers without relaying the data through a web server. But there are certain issues such as lack of full-fledged threading/concurrency support in the JavaScript language, reliance on synchronous loading etc. that restricts modern day browsers to take full advantage of current multiprocessing capabilities. Although, on one hand there are advantages of using web browsers, such as no requirement of specialized software, benefits of emerging technologies etc., the aforementioned issues pose challenges in implementation in certain areas. In this paper, we have tried to couple the benefits of peer-to-peer (P2P) architecture (elimination of centralized dependency, better scalability, shareability etc.) along with the advantages of recent web technologies (NodeJs, WebRTC etc.) by designing and implementing a browser based P2P content sharing framework. We have addressed the aforementioned challenges of a browser based P2P architecture by providing a mechanism to exchange messages asynchronously and facilitating new peer joins via existing peers in the network, thus reducing the dependency on bootstrap server. Our prototypical implementation demonstrates the feasibility, efficiency and scalability of this lightweight framework, on the top of which a variety of applications can be added as a layer of functionality.
A Browser-Based Distributed Framework for Content Sharing and Student Collaboration
(Springer, 2017-08) Haribabu, K; Sinha, Yash
The utilization of the networks in education system has become increasingly widespread in recent years. WebRTC has been one of the hottest topics recently when it comes to Web technologies for distributed systems as it enables peer-to-peer (P2P) connectivity between machines with higher reliability and better scalability without the overhead of resource management. In this paper, we propose a browser based, asynchronous framework of a P2P network using distributed, lookup protocol (Chord), NodeJS and RTCDataChannel; which is scalable and lightweight. The design combines the advantages of P2P networks for better and sophisticated education delivery. The framework will facilitate students to share course content and discuss with fellow students without requiring any centralized infrastructure support.
Comparative study of preprocessing and classification methods in character recognition of natural scene images
(Springer, 2025-01) Sinha, Yash
This paper presents an approach to character recognition in natural scene images. Recognizing such text is a challenging problem in the field of Computer Vision, more than the recognition of scanned documents due to several reasons. We propose a classification technique for classifying characters based on a pipeline of image processing operations and ensemble machine learning techniques. This pipeline tackles problems where Optical Character Recognition (OCR) fails. We present a framework that comprises a sequence of operations such as resizing, grey scaling, thresholding, morphological opening and median filtering on the images to handle background clutter, noise, multi-sized and multi-oriented characters and variance in illumination. We used image pixels and HOG (Histogram of Oriented Gradients) as features to train three different models based on Nearest-Neighbour, Random Forest and Extra Tree classifiers. When the input images were pre-processed, HOG features were extracted and fed into extra tree classifier, and the model classified the characters with maximum accuracy, among the other models that we tested. The proposed steps have been experimentally proven to yield better accuracy than the present state-of-the-art classification techniques on the Chars74k dataset. In addition, the paper includes a comparative study elaborating on various image processing operations, feature extraction methods and classification techniques.
Control-data plane intelligence trade-off in SDN
(IEEE, 2025) Sinha, Yash
With the decoupling of network control and data planes, the upcoming Software Defined Networking (SDN) paradigm advocates better network control and manageability. It introduces logical centralized control, network programmability and abstraction of underlying infrastructure from network services and applications. With global visibility of network state and central control that eases real time monitoring, policy alterations etc., it certainly enhances network security inherently. However, the separation of planes opens up new challenges like denial of service (DoS) attack, saturation attack, man-in-the middle attack and so on. Many of the issues of controller availability, controller-switch communication delay and scalability can be solved separately by distributed controllers, out-of-band communication links and parallelization respectively. Control-data plane intelligence trade-off has the potential to solve all of these. It increases controller availability, reduces latency for traffic engineering & decision making, and improves controller scalability. Moreover, control-data plane intelligence trade-off enables the control-data plane communication to be more secure. This will tremendously offload the processing load on the controller. We present how to realize control-data plane intelligence tradeoff extending OpenFlow.
Convolutional Neural Network-Based Human Identification Using Outer Ear Images
(Springer, 2018-10) Ajmera, Pawan K.; Sinha, Yash
This paper presents a deep learning approach for ear localization and recognition. The comparable complexity between human outer ear and face in terms of its uniqueness and permanence has increased interest in the use of ear as a biometric. But similar to face recognition, it poses challenges such as illumination, contrast, rotation, scale, and pose variation. Most of the techniques used for ear biometric authentication are based on traditional image processing techniques or handcrafted ensemble features. Owing to extensive work in the field of computer vision using convolutional neural networks (CNNs) and histogram of oriented gradients (HOG), the feasibility of deep neural networks (DNNs) in the field of ear biometrics has been explored in this research paper. A framework for ear localization and recognition is proposed that aims to reduce the pipeline for a biometric recognition system. The proposed framework uses HOG with support vector machines (SVMs) for ear localization and CNN for ear recognition. CNNs combine feature extraction and ear recognition tasks into one network with an aim to resolve issues such as variations in illumination, contrast, rotation, scale, and pose. The feasibility of the proposed technique has been evaluated on USTB III database. This work demonstrates 97.9% average recognition accuracy using CNNs without any image preprocessing, which shows that the proposed approach is promising in the field of biometric recognition.
Distill to delete: unlearning in graph networks with knowledge distillation
(2023-09) Sinha, Yash
Graph unlearning has emerged as a pivotal method to delete information from a pre-trained graph neural network (GNN). One may delete nodes, a class of nodes, edges, or a class of edges. An unlearning method enables the GNN model to comply with data protection regulations (i.e., the right to be forgotten), adapt to evolving data distributions, and reduce the GPU-hours carbon footprint by avoiding repetitive retraining. Existing partitioning and aggregation-based methods have limitations due to their poor handling of local graph dependencies and additional overhead costs. More recently, GNNDelete offered a model-agnostic approach that alleviates some of these issues. Our work takes a novel approach to address these challenges in graph unlearning through knowledge distillation, as it distills to delete in GNN (D2DGN). It is a model-agnostic distillation framework where the complete graph knowledge is divided and marked for retention and deletion. It performs distillation with response-based soft targets and feature-based node embedding while minimizing KL divergence. The unlearned model effectively removes the influence of deleted graph elements while preserving knowledge about the retained graph elements. D2DGN surpasses the performance of existing methods when evaluated on various real-world graph datasets by up to (AUC) in edge and node unlearning tasks. Other notable advantages include better efficiency, better performance in removing target elements, preservation of performance for the retained elements, and zero overhead costs. Notably, our D2DGN surpasses the state-of-the-art GNNDelete in AUC by , improves membership inference ratio by , requires fewer FLOPs per forward pass and up to faster.
Estimation of Raw Packets in SDN
(Springer, 2017-12) Haribabu, K; Sinha, Yash
In SDN based networks, for network management such as monitoring, performance tuning, enforcing security, configurations, calculating QoS metrics etc. a certain fraction of traffic is responsible. It consists of packets for many network protocols such as DHCP, MLD, MDNS, NDP etc. Most of the time these packets are created and absorbed at midway switches. We refer to these as raw packets. Cumulative statistics of sent and received traffic is sent to the controller by OpenFlow compliant switches that includes these raw packets. Although, not part of the data traffic these packets get counted and leads to noise in the measured statistics and thus, hamper the accuracy of methods that depend on these statistics such as calculation of QoS metrics. In this paper, we propose a method to estimate the fraction of the network traffic that consists of raw packets in Software Defined Networks. The number of raw packets transferred depends on the number of switches and hosts in the network and it is a periodic function of time. Through experiments on several network topologies, we have estimated a way to find a cap on the generated raw packets in the network, using spanning tree information about the topology.
Lung cancer detection: a deep learning approach
(Springer, 2018-10) Sinha, Yash
We present an approach to detect lung cancer from CT scans using deep residual learning. We delineate a pipeline of preprocessing techniques to highlight lung regions vulnerable to cancer and extract features using UNet and ResNet models. The feature set is fed into multiple classifiers, viz. XGBoost and Random Forest, and the individual predictions are ensembled to predict the likelihood of a CT scan being cancerous. The accuracy achieved is 84% on LIDC-IRDI outperforming previous attempts.
Meticulous Measurement of Control Packets in SDN
(ACM Digital Library, 2017-04) Haribabu, K; Sinha, Yash
The data packet statistics sent by OpenFlow compliant switches cumulatively includes statistics about control traffic which is used for network control and management. This reduces the accuracy of calculation of QoS metrics and thus hampers network monitoring. We present here a novel algorithm to accurately measure the fraction of control packets in SDN within 3% error rate.
MPLS based hybridization in SDN
(IEEE, 2017) Shekhawat, Virendra Singh; Chalapathi, G.S.S.; Sinha, Yash
The new paradigm of Software Defined Networking (SDN) although has great potential to address the complex problems presented by enterprise networks, it has its own deployment and scalability issues. Further, a full SDN deployment has its own business and economic challenges. A smooth transition from legacy networks to SDN (disruption free, accommodating budget constraints, with progressive improvement in network management) requires a hybrid networking model as an inevitable intermediate step; that allows heterogeneous paradigms to function together while the full transition is realized in phases. Therefore, the need of the hour is to develop an incremental deployment strategy that caters to the needs of the organization. We present here a class-based hybrid SDN model for Multi Protocol Label Switching (MPLS) networks. We discuss the model, design, components, their interactions, advantages and drawbacks. We also present a n implementation and evaluation of a prototype. In legacy networks, MPLS architecture closely resembles SDN paradigm in terms of separation of control and data planes, flow-abstraction etc. Moreover, ISPs have preferred MPLS over the years due to benefits of virtual private networks and traffic engineering. The central idea is to partition traffic using forwarding equivalence classes at the ingress router, the rules of which can be updated via a centralized controller using OpenFlow. Therefore, we aim to use the standard MPLS data-plane together with a control-plane based on OpenFlow to come up with a systematic incremental deployment methodology as well as a hybrid operation model
Multi-modal recommendation unlearning for legal, licensing, and modality constraints
(Association for the Advancement of Artificial Intelligence, 2025) Sinha, Yash
User data spread across multiple modalities has popularized multi-modal recommender systems (MMRS). They recommend diverse content such as products, social media posts, TikTok reels, etc., based on a user-item interaction graph. With rising data privacy demands, recent methods propose unlearning private user data from uni-modal recommender systems (RS). However, methods for unlearning item data related to outdated user preferences, revoked licenses, and legally requested removals are still largely unexplored. Previous RS unlearning methods are unsuitable for MMRS due to the incompatibility of their matrix-based representation with the multi-modal user-item interaction graph. Moreover, their data partitioning step degrades performance on each shard due to poor data heterogeneity and requires costly performance aggregation across shards. This paper introduces MMRecUn, the first approach known to us for unlearning in MMRS and unlearning item data. Given a trained RS model, MMRecUn employs a novel Reverse Bayesian Personalized Ranking (BPR) objective to enable the model to forget marked data. The reverse BPR attenuates the impact of user-item interactions within the forget set, while the forward BPR reinforces the significance of user-item interactions within the retain set. Our experiments demonstrate that MMRecUn outperforms baseline methods across various unlearning requests when evaluated on benchmark MMRS datasets. MMRecUn achieves recall performance improvements of up to 49.85% compared to baseline methods and is up to 1.3× faster than the Gold model, which is trained on retain set from scratch. MMRecUn offers significant advantages, including superiority in removing target interactions, preserving retained interactions, and zero overhead costs compared to previous methods.
Orgaccess: a benchmark for role based access control in organization scale llms
(2025-05) Sinha, Yash
Role-based access control (RBAC) and hierarchical structures are foundational to how information flows and decisions are made within virtually all organizations. As the potential of Large Language Models (LLMs) to serve as unified knowledge repositories and intelligent assistants in enterprise settings becomes increasingly apparent, a critical, yet under explored, challenge emerges: \textit{can these models reliably understand and operate within the complex, often nuanced, constraints imposed by organizational hierarchies and associated permissions?} Evaluating this crucial capability is inherently difficult due to the proprietary and sensitive nature of real-world corporate data and access control policies. We introduce a synthetic yet representative \textbf{OrgAccess} benchmark consisting of 40 distinct types of permissions commonly relevant across different organizational roles and levels. We further create three types of permissions: 40,000 easy (1 permission), 10,000 medium (3-permissions tuple), and 20,000 hard (5-permissions tuple) to test LLMs' ability to accurately assess these permissions and generate responses that strictly adhere to the specified hierarchical rules, particularly in scenarios involving users with overlapping or conflicting permissions. Our findings reveal that even state-of-the-art LLMs struggle significantly to maintain compliance with role-based structures, even with explicit instructions, with their performance degrades further when navigating interactions involving two or more conflicting permissions. Specifically, even \textbf{GPT-4.1 only achieves an F1-Score of 0.27 on our hardest benchmark}. This demonstrates a critical limitation in LLMs' complex rule following and compositional reasoning capabilities beyond standard factual or STEM-based benchmarks, opening up a new paradigm for evaluating their fitness for practical, structured environments.
Real Time Monitoring of Packet Loss in Software Defined Networks
(Springer, 2017-12) Haribabu, K; Sinha, Yash
In order to meet QoS demands from customers, currently, ISPs over-provision capacity. Networks need to continuously monitor performance metrics, such as bandwidth, packet loss etc., in order to quickly adapt forwarding rules in response to changes in the workload. The packet loss metric is also required by network administrators and ISPs to identify clusters in network that are vulnerable to congestion. However, the existing solutions either require special instrumentation of the network or impose significant measurement overhead. Software-Defined Networking (SDN), an emerging paradigm in networking advocates separation of the data plane and the control plane, separating the network’s control logic from the underlying routers and switches, leaving a logically centralized software program to control the behavior of the entire network, and introducing network programmability. Further, OpenFlow allows to implement fine-grained Traffic Engineering (TE) and provides flexibility to determine and enforce end-to-end QoS parameters.
Step-by-step reasoning attack: revealing 'erased' knowledge in large language models
(2025-06) Sinha, Yash
Knowledge erasure in large language models (LLMs) is important for ensuring compliance with data and AI regulations, safeguarding user privacy, mitigating bias, and misinformation. Existing unlearning methods aim to make the process of knowledge erasure more efficient and effective by removing specific knowledge while preserving overall model performance, especially for retained information. However, it has been observed that the unlearning techniques tend to suppress and leave the knowledge beneath the surface, thus making it retrievable with the right prompts. In this work, we demonstrate that \textit{step-by-step reasoning} can serve as a backdoor to recover this hidden information. We introduce a step-by-step reasoning-based black-box attack, Sleek, that systematically exposes unlearning failures. We employ a structured attack framework with three core components: (1) an adversarial prompt generation strategy leveraging step-by-step reasoning built from LLM-generated queries, (2) an attack mechanism that successfully recalls erased content, and exposes unfair suppression of knowledge intended for retention and (3) a categorization of prompts as direct, indirect, and implied, to identify which query types most effectively exploit unlearning weaknesses. Through extensive evaluations on four state-of-the-art unlearning techniques and two widely used LLMs, we show that existing approaches fail to ensure reliable knowledge removal. Of the generated adversarial prompts, 62.5% successfully retrieved forgotten Harry Potter facts from WHP-unlearned Llama, while 50% exposed unfair suppression of retained knowledge. Our work highlights the persistent risks of information leakage, emphasizing the need for more robust unlearning strategies for erasure.
Studying the role of kinect as a multi-sensory learning platform for children
(ACM Digital Library, 2018) Sinha, Yash
According to the theory of Embodied Cognition, our behavior is a result of real-time interaction with surroundings, our cognitive skills, and the nervous system. From this perspective, researchers are considering a learning environment which promotes physical activities to achieve cognitive tasks. Such Natural User Interfaces (NUI) make use of gesture-based sensors like the Microsoft Kinect. Yet we lack in-depth studies of how they improve the learning process. In this paper, we present observations of two deployment studies which focus on different roles that NUI can play as a part of learning activities. We deploy the Kinect based applications:- Yoga Soft: A Digital Yoga Instructor and Mudra: A Kinect based Learning System in real life scenarios. The first study is conducted at residences of preadolescent children in Gurgaon, India. The second study is conducted at an education center specializing in the care of kindergarten children in Pilani, India.
A survey: Hybrid SDN
(Elsevier, 2017-12) Haribabu, K; Sinha, Yash
A full deployment of Software Defined Networking (SDN) paradigm poses multi-dimensional challenges viz., technical, financial and business challenges. Technical challenges of scalability, fault tolerance, centralization guarantees exist. Financial challenges of budget constraints, non-availability of phased transition model exist. Business challenges like acceptability, building confidence among network operators etc. exist. Therefore, a direct and sudden transition from legacy networks to pure SDN seems unlikely. A hybrid deployment of SDN can be one of the plausible intermediate paths, primarily because it provides an environment where both legacy and SDN nodes can work together. Thus, an incremental deployment strategy can be developed. Further, hybrid SDN can enforce the benefits of both the traditional networks and SDN paradigm. Hybrid SDN deployment has many advantages including adaptability to budget constraints, central programmability of the network, fallback to time-tested legacy mechanisms and so on. But there are challenges specific to hybrid models, like added complexity of running multiple paradigms together, realizing cooperation between control planes, etc. We envision that more research work is needed to maximize the benefits and limit the drawbacks.
Towards robust concept erasure in diffusion models: unlearning identity, nudity and artistic styles
(2024-11) Sinha, Yash
Diffusion models have achieved remarkable success in generative tasks across various domains. However, the increasing demand for content moderation and the removal of specific concepts from these models has introduced the challenge of \textit{unlearning}. In this work, we present a suite of robust methodologies that significantly enhance the unlearning process by employing advanced loss functions within knowledge distillation frameworks. Specifically, we utilize the Cramer-Wold distance and Jensen-Shannon (JS) divergence to facilitate more efficient and versatile concept removal. Although current non-learning techniques are effective in certain scenarios, they are typically limited to specific categories such as identity, nudity, or artistic style. In contrast, our proposed methods demonstrate robust versatility, seamlessly adapting to and performing effectively across a wide range of concept erasure categories. Our approach outperforms existing techniques, achieving consistent results across different unlearning categories and showcasing its broad applicability. Through extensive experiments, we show that our method not only surpasses previous benchmarks but also addresses key limitations of current unlearning techniques, paving the way for more responsible use of text-to-image diffusion models.
Unstar: unlearning with self-taught anti-sample reasoning for ILMS
(2024-10) Sinha, Yash
The key components of machine learning are data samples for training, model for learning patterns, and loss function for optimizing accuracy. Analogously, unlearning can potentially be achieved through anti-data samples (or anti-samples), unlearning method, and reversed loss function. While prior research has explored unlearning methods and reversed loss functions, the potential of anti-samples remains largely untapped. In this paper, we introduce UnSTAR: Unlearning with Self-Taught Anti-Sample Reasoning for large language models (LLMs). Our contributions are threefold; first, we propose a novel concept of anti-sample-induced unlearning; second, we generate anti-samples by leveraging misleading rationales, which help reverse learned associations and accelerate the unlearning process; and third, we enable fine-grained targeted unlearning, allowing for the selective removal of specific associations without impacting related knowledge - something not achievable by previous works. Results demonstrate that anti-samples offer an efficient, targeted unlearning strategy for LLMs, opening new avenues for privacy-preserving machine learning and model modification.