arXiv subjects

Jeya Pradha Jeyaraj

Publications and source records attributed to Jeya Pradha Jeyaraj.

4 recordsLinked to original sources

arXiv2026

Video Reasoning without Training

Video reasoning using Large Multimodal Models (LMMs) relies on costly reinforcement learning (RL) and verbose chain-of-thought, resulting in substantial computational overhead during both training and inference. Moreover, the mechanisms that control the thinking process in these reasoning models are very limited. In this paper, we use the entropy of the model's output distribution as a signal to study and guide reasoning behavior. We discover that high-quality models exhibit a characteristic pattern of micro-exploration and micro-exploitation cycles, followed by a later entropy peak (i.e., longer thinking) and a lower final entropy, indicating more deliberate exploration and confident convergence (i.e., avoid excessive randomness while the model is exploring or thinking through an answer). We then use these novel, theoretically-grounded insights to introduce V-Reason (Video-Reason), an inference-time optimization method that adapts the value cache of the LMM through a lightweight, trainable controller. Our proposed controller is guided by an entropy-based objective, to tune the model's behavior directly at inference, without using any RL or supervised fine-tuning. Our experiments show that V-Reason significantly outperforms the base instruction-tuned models on many video reasoning datasets, narrowing the gap with RL models to within 0.6% accuracy on average. We achieve this without any training, while offering efficiency benefits: V-Reason uses 58.6% fewer tokens than the RL model. Project Page https://deepaksridhar.github.io/vreason.github.io/

cs.CV↗

arXiv2021

The Transdimensional Poisson Process for Vehicular Network Analysis

A comprehensive vehicular network analysis requires modeling the street system and vehicle locations. Even when Poisson point processes (PPPs) are used to model the vehicle locations on each street, the analysis is barely tractable. That holds for even a simple average-based performance metric -- the success probability, which is a special case of the fine-grained metric, the meta distribution (MD) of the signal-to-interference ratio (SIR). To address this issue, we propose the transdimensional approach as an alternative. Here, the union of 1D PPPs on the streets is simplified to the transdimensional PPP (TPPP), a superposition of 1D and 2D PPPs. The TPPP includes the 1D PPPs on the streets passing through the receiving vehicle and models the remaining vehicles as a 2D PPP ignoring their street geometry. Through the SIR MD analysis, we show that the TPPP provides good approximations to the more cumbrous models with streets characterized by Poisson line/stick processes; and we prove that the accuracy of the TPPP further improves under shadowing. Lastly, we use the MD results to control network congestion by adjusting the transmit rate while maintaining a target fraction of reliable links. A key insight is that the success probability is an inadequate measure of congestion as it does not capture the reliabilities of the individual links.

cs.NI↗

arXiv2021

Cox Models for Vehicular Networks: SIR Performance and Equivalence

We introduce a general framework for the modeling and analysis of vehicular networks by defining street systems as random 1D subsets of $\mathbb{R}^{2}$. The street system, in turn, specifies the random intensity measure of a Cox process of vehicles, i.e., vehicles form independent 1D Poisson point processes on each street. Models in this Coxian framework can characterize streets of different lengths and orientations forming intersections or T-junctions. The lengths of the streets can be infinite or finite and mutually independent or dependent. We analyze the reliability of communication for different models, where reliability is the probability that a vehicle at an intersection, a T-junction, or a general location can receive a message successfully from a transmitter at a certain distance. Further, we introduce a notion of equivalence between vehicular models, which means that a representative model can be used as a proxy for other models in terms of reliability. Specifically, we prove that the Poisson stick process-based vehicular network is equivalent to the Poisson line process-based and Poisson lilypond model-based vehicular networks, and their rotational variants.

cs.NI↗

arXiv2016

Resource Allocation and Fairness in Wireless Powered Cooperative Cognitive Radio Networks

We integrate a wireless powered communication network with a cooperative cognitive radio network, where multiple secondary users (SUs) powered wirelessly by a hybrid access point (HAP) help a primary user relay the data. As a reward for the cooperation, the secondary network gains the spectrum access where SUs transmit to HAP using time division multiple access. To maximize the sum-throughput of SUs, we present a secondary sum-throughput optimal resource allocation (STORA) scheme. Under the constraint of meeting target primary rate, the STORA scheme chooses the optimal set of relaying SUs and jointly performs the time and energy allocation for SUs. Specifically, by exploiting the structure of the optimal solution, we find the order in which SUs are prioritized to relay primary data. Since the STORA scheme focuses on the sum-throughput, it becomes inconsiderate towards individual SU throughput, resulting in low fairness. To enhance fairness, we investigate three resource allocation schemes, which are (i) equal time allocation, (ii) minimum throughput maximization, and (iii) proportional time allocation. Simulation results reveal the trade-off between sum-throughput and fairness. The minimum throughput maximization scheme is the fairest one as each SU gets the same throughput, but yields the least SU sum-throughput.

cs.IT↗