arXiv subjects

Dipak Ghosal

Publications and source records attributed to Dipak Ghosal.

5 recordsLinked to original sources

arXiv2025

Enhancing Japanese Large Language Models with Reasoning Vectors

Post-training methods have improved the performance and enhanced the reasoning capability for mainstream large language models (LLMs), but the same is challenging for Japanese LLMs to achieve due to the amount of resources required. Inspired by task vectors that extract the change of weights before and after training, specifically for a certain task, we obtain reasoning vectors from reasoning LLMs and apply them to Japanese LLMs to boost their performance. While the resources available present a challenge to improve Japanese LLMs, we present a simple and effective way to obtain high improvement and hope to inspire for other languages.

cs.AI↗

arXiv2024

CourseAssist: Pedagogically Appropriate AI Tutor for Computer Science Education

The growing enrollments in computer science courses and increase in class sizes necessitate scalable, automated tutoring solutions to adequately support student learning. While Large Language Models (LLMs) like GPT-4 have demonstrated potential in assisting students through question-answering, educators express concerns over student overreliance, miscomprehension of generated code, and the risk of inaccurate answers. Rather than banning these tools outright, we advocate for a constructive approach that harnesses the capabilities of AI while mitigating potential risks. This poster introduces CourseAssist, a novel LLM-based tutoring system tailored for computer science education. Unlike generic LLM systems, CourseAssist uses retrieval-augmented generation, user intent classification, and question decomposition to align AI responses with specific course materials and learning objectives, thereby ensuring pedagogical appropriateness of LLMs in educational settings. We evaluated CourseAssist against a baseline of GPT-4 using a dataset of 50 question-answer pairs from a programming languages course, focusing on the criteria of usefulness, accuracy, and pedagogical appropriateness. Evaluation results show that CourseAssist significantly outperforms the baseline, demonstrating its potential to serve as an effective learning assistant. We have also deployed CourseAssist in 6 computer science courses at a large public R1 research university reaching over 500 students. Interviews with 20 student users show that CourseAssist improves computer science instruction by increasing the accessibility of course-specific tutoring help and shortening the feedback loop on their programming assignments. Future work will include extensive pilot testing at more universities and exploring better collaborative relationships between students, educators, and AI that improve computer science learning experiences.

cs.CY↗

arXiv2020

TCP D*: A Low Latency First Congestion Control Algorithm

The choice of feedback mechanism between delay and packet loss has long been a point of contention in TCP congestion control. This has partly been resolved, as it has become increasingly evident that delay based methods are needed to facilitate modern interactive web applications. However, what has not been resolved is what control should be used, with the two candidates being the congestion window and the pacing rate. BBR is a new delay based congestion control algorithm that uses a pacing rate as its primary control and the congestion window as a secondary control. We propose that a congestion window first algorithm might give more desirable performance characteristics in situations where latency must be minimized even at the expense of some loss in throughput. To evaluate this hypothesis we introduce a new congestion control algorithm called TCP D*, which is a congestion window first algorithm that adopts BBR's approach of maximizing delivery rate while minimizing latency. In this paper, we discuss the key features of this algorithm, discuss the differences and similarity to BBR, and present some preliminary results based on a real implementation.

cs.NI↗

arXiv2020

Model Predictive Congestion Control for TCP Endpoints

A common problem in science networks and private wide area networks (WANs) is that of achieving predictable data transfers of multiple concurrent flows by maintaining specific pacing rates for each. We address this problem by developing a control algorithm based on concepts from model predictive control (MPC) to produce flows with smooth pacing rates and round trip times (RTTs). In the proposed approach, we model the bottleneck link as a queue and derive a model relating the pacing rate and the RTT. A MPC based control algorithm based on this model is shown to avoid the extreme window (which translates to rate) reduction that exists in current control algorithms when facing network congestion. We have implemented our algorithm as a Linux kernel module. Through simulation and experimental analysis, we show that our algorithm achieves the goals of a low standard deviation of RTT and pacing rate, even when the bottleneck link is fully utilized. In the case of multiple flows, we can assign different rates to each flow and as long as the sum of rates is less than bottleneck rate, they can maintain their assigned pacing rate with low standard deviation. This is achieved even when the flows have different RTTs.

cs.NI↗

arXiv2018

A Survey of Multimedia Streaming in LTE Cellular Networks

With the growing of Long Term Evolution (LTE) cellular networks and the increase in the demand of the video services, it is vital to consider the challenges in the streaming services from a different perspective. A perspective that focuses on the streaming services in light of cellular networks challenges, both per layer basis and across multiple layers as well. In this tutorial, we highlight the main challenges that faces the industry of video streaming in the context of cellular networks with a focus on LTE. We also discuss proposed solutions for these challenges while highlighting the limitations of these solutions and the conditions/assumptions required for these solution to deliver high performance. In addition, we show different work in cross layer optimization for video streaming and how it leads towards a more optimized end to end LTE networking for video streaming. Finally, we suggest different open research areas in the domain of video delivery over LTE networks that can significantly enhance the quality of streaming experience to the end user.

cs.NI↗