arXiv subjects

Ruchir Gupta

Publications and source records attributed to Ruchir Gupta.

10 recordsLinked to original sources

arXiv2026

HiMed: Incentivizing Hindi Reasoning in Medical LLMs

Medical large language models hold promise for reducing healthcare disparities, yet Hindi remains severely underrepresented. While medical LLMs excel in high-resource languages, their performance degrades sharply in Hindi, particularly on Indian systems of medicine. We argue that robust cross-lingual medical transfer requires Hindi reasoning. To this end, we introduce HiMed, a Hindi reasoning medical corpus and benchmark suite covering both Western and Indian medicine. We further propose HiMed-8B, a Hindi-form medical reasoning LLM, through the design of decaying scaffolding reward. Extensive experiments demonstrate improvement in Hindi medical reasoning performance and reduction in the English--Hindi accuracy gap. Ablation studies validate the contribution of each training stage and reward component. All data and code are available on GitHub: https://github.com/FreedomIntelligence/HiMed.

cs.CL↗

arXiv2017

Evolutionary Stability of Reputation Management System in Peer to Peer Networks

Each participant in peer-to-peer network prefers to free-ride on the contribution of other participants. Reputation based resource sharing is a way to control the free riding. Instead of classical game theory we use evolutionary game theory to analyse the reputation based resource sharing in peer to peer system. Classical game-theoretical approach requires global information of the population. However, the evolutionary games only assumes light cognitive capabilities of users, that is, each user imitates the behavior of other user with better payoff. We find that without any extra benefit reputation strategy is not stable in the system. We also find the fraction of users who calculate the reputation for controlling the free riding in equilibrium. In this work first we made a game theoretical model for the reputation system and then we calculate the threshold of the fraction of users with which the reputation strategy is sustainable in the system. We found that in simplistic conditions reputation calculation is not evolutionarily stable strategy but if we impose some initial payment to all users and then distribute that payment among the users who are calculating reputation then reputation is evolutionary stable strategy.

cs.GT↗

arXiv2016

A Parameter-free Affinity Based Clustering

Several methods have been proposed to estimate the number of clusters in a dataset; the basic ideal behind all of them has been to study an index that measures inter-cluster separation and intra-cluster cohesion over a range of cluster numbers and report the number which gives an optimum value of the index. In this paper we propose a simple, parameter free approach that is like human cognition to form clusters, where closely lying points are easily identified to form a cluster and total number of clusters are revealed. To identify closely lying points, affinity of two points is defined as a function of distance and a threshold affinity is identified, above which two points in a dataset are likely to be in the same cluster. Well separated clusters are identified even in the presence of outliers, whereas for not so well separated dataset, final number of clusters are estimated and the detected clusters are merged to produce the final clusters. Experiments performed with several large dimensional synthetic and real datasets show good results with robustness to noise and density variation within dataset.

cs.CV↗

arXiv2015

Resource allocation in Peer-to-Peer Networks: A Control-Theoretical Perspective

P2P system rely on voluntary allocation of resources by its members due to absence of any central controlling authority. This resource allocation can be viewed as classical control problem where feedback is the amount of resource received, which controls the output i.e. the amount of resources shared back to the network by the node. The motivation behind the use of control system in resource allocation is to exploit already existing tools in control theory to improve the overall allocation process and thereby solving the problem of freeriding and whitewashing in the network. At the outset, we have derived the transfer function to model the P2P system. Subsequently, through the simulation results we have shown that transfer function was able to provide optimal value of resource sharing for the peers during the normal as well as high degree of overloading in the network. Thereafter we verified the accuracy of the transfer function derived by comparing its output with the simulated P2P network. To demonstrate how control system reduces free riding it has been shown through simulations how the control systems penalizes the nodes indulging in different levels of freeriding. Our proposed control system shows considerable gain over existing state of art algorithm. This improvement is achieved through PI action of controller. Since low reputation peers usually subvert reputation system by whitewashing. We propose and substantiate a technique modifying transfer function such that systems' sluggishness becomes adaptive in such a way that it encourage genuine new comers to enter network and discourages member peers to whitewash.

cs.NI↗

arXiv2015

An Efficient Local Strategy to Control Information Spreading in Network

In social networks, control of rumor spread is an active area of research. SIR model is generally used to study the rumor dynamics in network while considering the rumor as an epidemic. In disease spreading model, epidemic is controlled by removing central nodes in the network. Full network information is needed for such removal. To have the information of complete network is difficult proposition. As a consequence, the search of an algorithm that may control epidemic without needing global information is a matter of great interest. In this paper, an immunization strategy is proposed that uses only local information available at a node, viz. degree of the node and average degree of its neighbour nodes. Proposed algorithm has been evaluated for scale-free network using SIR model. Numerical results show that proposed method has less complexity and gives significantly better results in comparison with other strategies while using only local information.

cs.SI↗

arXiv2014

Reputation Aggregation in Peer-to-Peer Network Using Differential Gossip Algorithm

Reputation aggregation in peer to peer networks is generally a very time and resource consuming process. Moreover, most of the methods consider that a node will have same reputation with all the nodes in the network, which is not true. This paper proposes a reputation aggregation algorithm that uses a variant of gossip algorithm called differential gossip. In this paper, estimate of reputation is considered to be having two parts, one common component which is same with every node, and the other one is information received from immediate neighbours based on the neighbours' direct interaction with the node. The differential gossip is fast and requires less amount of resources. This mechanism allows computation of independent reputation value by a node, of every other node in the network, for each node. The differential gossip trust has been investigated for a power law network formed using preferential attachment \emph{(PA)} Model. The reputation computed using differential gossip trust shows good amount of immunity to the collusion. We have verified the performance of the algorithm on the power law networks of different sizes ranging from 100 nodes to 50,000 nodes.

cs.NI↗

arXiv2013

Adaptive Push-Then-Pull Gossip Algorithm for Scale-free Networks

Real life networks are generally modelled as scale free networks. Information diffusion in such networks in decentralised environment is a difficult and resource consuming affair. Gossip algorithms have come up as a good solution to this problem. In this paper, we have proposed Adaptive First Push Then Pull gossip algorithm. We show that algorithm works with minimum cost when the transition round to switch from Adaptive Push to Adaptive Pull is close to Round(log(N)). Furthermore, we compare our algorithm with Push, Pull and First Push Then Pull and show that the proposed algorithm is the most cost efficient in Scale Free networks.

cs.NI↗

arXiv2013

A Reputation Based Framework to Avoid Free-riding in Unstructured Peer-to-Peer network

Free riding is a major problem in peer-to-peer networks. Reputation management systems are generally proposed to overcome this problem. In this paper we have discussed a possible way of resource allocation on the basis of reputation management system i.e. probabilistic allocation based on reputation. This seems to be a better way for allocation of resources because in this case nodes that do not have very good reputation about each other, may also serve each other at least some amount of resource with finite probability. This avoids disconnect between them. Algorithms are presented for optimizing the shared capacity, reputation based probabilistic allocation that is optimal for a node, and formation of interest groups on the basis of similarity between interests of nodes.

cs.NI↗

arXiv2013

Avoiding Whitewashing in Unstructured Peer-to-Peer Resource Sharing Network

In peer-to-peer file sharing network, it is hard to distinguish between a legitimate newcomer and a whitewasher. This makes whitewashing a big problem in peer-to-peer networks. Although the problem of whitewashing can be solved using permanent identities, it may take away the right of anonymity for users. In this paper, we a have proposed a novel algorithm to avoid this problem when network uses free temporary identities. In this algorithm, the initial reputation is adjusted according to the level of whitewashing in the network.

cs.NI↗

arXiv2013

Trust Estimation in Peer-to-Peer Network Using BLUE

In peer-to-peer networks, free riding is a major problem. Reputation management systems can be used to overcome this problem. Reputation estimation methods generally do not consider the uncertainties in the inputs. We propose a reputation estimation method using BLUE (Best Linear Unbiased estimator) estimator that consider uncertainties in the input variables.

cs.NI↗