arXiv subjects

Kasthuri Jayarajah

Publications and source records attributed to Kasthuri Jayarajah.

7 recordsLinked to original sources

arXiv2026

Dynamic Conformance Testing of WebGPU Through Specification-Driven Mutation

WebGPU is a low-level graphics and compute API that exposes modern GPU functionality to web applications. While the official WebGPU Conformance Test Suite (CTS) focuses on well-formed usage under the WebGPU specification, it is not designed to stress implementations with semantic edge cases or adversarial inputs. General-purpose fuzzers, in contrast, struggle with WebGPU because of its complex graphics stack and multi-process architecture. We introduce LANTERN, a specification-guided dynamic conformance testing framework that mutates CTS tests using constraints extracted from the WebGPU specification. LANTERN extracts explicit syntactic API rules from WebIDL definitions and recovers semantic constraints, such as command ordering and object lifetimes, from natural-language specification text. Selected rules guide AST-located textual transformations that generate both valid and intentionally invalid CTS variants. We execute the resulting tests at scale on AddressSanitizer-instrumented Chromium to discover bugs. Our evaluation discovers three reproducible bugs, including a heap corruption, in Chromium versions current at the time of study. These results demonstrate that syntactically seeded, semantics-aware mutation of conformance tests provides a way to uncover browser bugs during WebGPU testing.

cs.SE↗

arXiv2025

Memento: Augmenting Personalized Memory via Practical Multimodal Wearable Sensing in Visual Search and Wayfinding Navigation

Working memory involves the temporary retention of information over short periods. It is a critical cognitive function that enables humans to perform various online processing tasks, such as dialing a phone number, recalling misplaced items' locations, or navigating through a store. However, inherent limitations in an individual's capacity to retain information often result in forgetting important details during such tasks. Although previous research has successfully utilized wearable and assistive technologies to enhance long-term memory functions (e.g., episodic memory), their application to supporting short-term recall in daily activities remains underexplored. To address this gap, we present Memento, a framework that uses multimodal wearable sensor data to detect significant changes in cognitive state and provide intelligent in situ cues to enhance recall. Through two user studies involving 15 and 25 participants in visual search navigation tasks, we demonstrate that participants receiving visual cues from Memento achieved significantly better route recall, improving approximately 20-23% compared to free recall. Furthermore, Memento reduced cognitive load and review time by 46% while also substantially reducing computation time (3.86 seconds vs. 15.35 seconds), offering an average of 75% effectiveness compared to computer vision-based cue selection approaches.

cs.HC↗

arXiv2022

TagTeam: Towards Wearable-Assisted, Implicit Guidance for Human--Drone Teams

The availability of sensor-rich smart wearables and tiny, yet capable, unmanned vehicles such as nano quadcopters, opens up opportunities for a novel class of highly interactive, attention-shared human--machine teams. Reliable, lightweight, yet passive exchange of intent, data and inferences within such human--machine teams make them suitable for scenarios such as search-and-rescue with significantly improved performance in terms of speed, accuracy and semantic awareness. In this paper, we articulate a vision for such human--drone teams and key technical capabilities such teams must encompass. We present TagTeam, an early prototype of such a team and share promising demonstration of a key capability (i.e., motion awareness).

cs.HC↗

arXiv2020

Enabling Collaborative Video Sensing at the Edge through Convolutional Sharing

While Deep Neural Network (DNN) models have provided remarkable advances in machine vision capabilities, their high computational complexity and model sizes present a formidable roadblock to deployment in AIoT-based sensing applications. In this paper, we propose a novel paradigm by which peer nodes in a network can collaborate to improve their accuracy on person detection, an exemplar machine vision task. The proposed methodology requires no re-training of the DNNs and incurs minimal processing latency as it extracts scene summaries from the collaborators and injects back into DNNs of the reference cameras, on-the-fly. Early results show promise with improvements in recall as high as 10% with a single collaborator, on benchmark datasets.

cs.CV↗

arXiv2019

Inferring Accurate Bus Trajectories from Noisy Estimated Arrival Time Records

Urban commuting data has long been a vital source of understanding population mobility behaviour and has been widely adopted for various applications such as transport infrastructure planning and urban anomaly detection. While individual-specific transaction records (such as smart card (tap-in, tap-out) data or taxi trip records) hold a wealth of information, these are often private data available only to the service provider (e.g., taxicab operator). In this work, we explore the utility in harnessing publicly available, albeit noisy, transportation datasets, such as noisy "Estimated Time of Arrival" (ETA) records (commonly available to commuters through transit Apps or electronic signages). We first propose a framework to extract accurate individual bus trajectories from such ETA records, and present results from both a primary city (Singapore) and a secondary city (London) to validate the techniques. Finally, we quantify the upper bound on the spatiotemporal resolution, of the reconstructed trajectory outputs, achieved by our proposed technique.

cs.OH↗

arXiv2019

BuSCOPE : Fusing Individual & Aggregated Mobility Behavior for "Live" Smart City Services

While analysis of urban commuting data has a long and demonstrated history of providing useful insights into human mobility behavior, such analysis has been performed largely in offline fashion and to aid medium-to-long term urban planning. In this work, we demonstrate the power of applying predictive analytics on real-time mobility data, specifically the smart-card generated trip data of millions of public bus commuters in Singapore, to create two novel and "live" smart city services. The key analytical novelty in our work lies in combining two aspects of urban mobility: (a) conformity: which reflects the predictability in the aggregated flow of commuters along bus routes, and (b) regularity: which captures the repeated trip patterns of each individual commuter. We demonstrate that the fusion of these two measures of behavior can be performed at city-scale using our BuScope platform, and can be used to create two innovative smart city applications. The Last-Mile Demand Generator provides O(mins) lookahead into the number of disembarking passengers at neighborhood bus stops; it achieves over 85% accuracy in predicting such disembarkations by an ingenious combination of individual-level regularity with aggregate-level conformity. By moving driverless vehicles proactively to match this predicted demand, we can reduce wait times for disembarking passengers by over 75%. Independently, the Neighborhood Event Detector uses outlier measures of currently operating buses to detect and spatiotemporally localize dynamic urban events, as much as 1.5 hours in advance, with a localization error of 450 meters.

physics.soc-ph↗

arXiv2019

Can multimodal sensing detect and localize transient events?

With the increased focus on making cities "smarter", we see an upsurge in investment in sensing technologies embedded in the urban infrastructure. The deployment of GPS sensors aboard taxis and buses, smartcards replacing paper tickets, and other similar initiatives have led to an abundance of data on human mobility, generated at scale and available real-time. Further still, users of social media platforms such as Twitter and LBSNs continue to voluntarily share multimedia content revealing in-situ information on their respective localities. The availability of such longitudinal multimodal data not only allows for both the characterization of the dynamics of the city, but also, in detecting anomalies, resulting from events (e.g., concerts) that disrupt such dynamics, transiently. In this work, we investigate the capabilities of such urban sensor modalities, both physical and social, in detecting a variety of local events of varying intensities (e.g., concerts) using statistical outlier detection techniques. We look at loading levels on arriving bus stops, telecommunication records, and taxi trips, accrued via the public APIs made available through the local transport authorities from Singapore and New York City, and Twitter/Foursquare check-ins collected during the same period, and evaluate against a set of events assimilated from multiple event websites. In particular, we report on our early findings on (1) the spatial impact evident via each modality (i.e., how far from the event venue is the anomaly still present), and (2) the utility in combining decisions from the collection of sensors using rudimentary fusion techniques.

eess.SP↗