Kishan Maharaj, Vitobha Munigala, Srikanth G. Tamilselvam, Prince Kumar, Sayandeep Sen, Palani Kodeswaran, Abhijit Mishra, Pushpak Bhattacharyya. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Containerized Network Functions (CNFs) have seen recent increase in adoption due to the success of orchestration platforms and the natural flexibility of containers for deployment to heterogeneous, resource constrained environments like the edge. However, the weak isolation model of containers makes them susceptible to performance degradation due to interference from other containers. The interference manifests as stress on various resources like CPU, memory, cache, bandwidth of a CNF. While identifying the exact resource that is under stress is essential to employ the right remediation strategies, it is a challenging problem given the wide spectrum of applications, platforms and hardware, especially with the high performance requirements of CNFs with stringent SLAs. We present Sari, a first of a kind practical NF -agnostic framework leveraging temporal patterns in syscalls. Sari works for multi-service NFs and generalizes to a wide range of conditions. We leverage supervised time series classification and achieve performance upto 98.9% with minimal captures at runtime (starting from 25ms). Sari'sprediction error is 63.33% lower than any existing stress identification framework for VNFs and to the best of our knowledge, the first to address the problem for CNFs.

A Fair Digital Exchange is defined as either all or none of the participants achieving a (predetermined) desirable outcome. This work addresses third party mediated systems for digital content where mutually unknown, and hence non-trusting, buyers, sellers and the mediator (third party) take part in an exchange protocol. We address the lack of guaranteed fairness, as defined above, in the existing platforms for this setting. We present TEDX, a decentralized solution for guaranteed three party fair exchange of digital goods with scalability and support for incremental deployment over the existing (non-fair) platforms. TEDX combines carefully crafted message exchanges with incentive schemes designed to deter malicious behavior. TEDX also leverages ideas from blockchain anchored state-channels to provide trusted execution while minimizing the operational overheads of blockchain. We present the design and a security analysis of TEDX to validate the claimed fairness properties. We also present the details of a prototype implementation of TEDX leveraging Hyperledger Fabric and performance evaluation of the same on a realistic testbed spanning five public cloud zones. Our results indicate that TEDX adds only a minimal overhead of 16% while being 46x faster than a naive blockchain solution, thereby demonstrating that TEDX is scalable.

With the recent advances in Internet of Things, smarthomes containing a variety of sensors are expected to monitor and react to the activities of daily living (ADL) of the residents with the goal of improving convenience, comfort, and safety. However, a single human activity may trigger multiple sensors, and each sensor is triggered by multiple activities. Also, the very human nature of interleaving activities along with the presence of multiple inhabitants makes ADL detection a complex problem. This paper tackles the problem with a data driven de-multiplexing approach termed TSFS (Temporal-Sensor-Frequency-Stitch) that disentangles each individual activity from the sensor stream and thus simplifies the ADL recognition problem. TSFS leverages temporal(T), sensory(S), and frequency(F) information to discriminate between two consecutive/parallel activities, followed by a stitch(S) to enhance segmentation. The multi-faceted approach helps us in correctly identifying segments; extensive evaluation on 10 public datasets reveals that TSFS is better by 185 percent when compared to a simple temporal segmentation approach. The activity detection accuracy yields within 93 percent of the best possible (an oracle) approach.

Powered by the emergence of the Internet of Things, smart homes containing a variety of sensors and actuators are expected to monitor and react to the activities of the residents with the goal of improving convenience, comfort and safety. However, in typical home settings, each human Activity of Daily Living (ADL) generates events from multiple sensors, and each sensor is triggered by multiple ADLs. Consequently, achieving high detection accuracy in these complex environments requires large amounts of training data for every possible multiplexing scenario, making it a complex problem. In this paper, we propose a data driven three-step de-multiplexing approach that simplifies the ADL recognition problem by first segmenting the event stream into periods of interest, before feeding to a classifier. We mine datasets to identify salient features which allow us to achieve a good segmentation. Extensive evaluation on ten public datasets shows that our approach achieves upto 77% segmentation accuracy, and a activity detection accuracy within 91% of the best possible.

IoT enabled smart environments are expected to proliferate significantly in the near future, particularly in the context of monitoring services for wellness living, patient healthcare and elderly care. Timely maintenance of failed sensors is of critical importance in such deployments to ensure minimal disruption to monitoring services. However, maintenance of large and geographically spread deployments can be a significant challenge. We present Idea that significantly increases the vtime-before-repair for a smart home deployment, thereby reducing the maintenance overhead. Specifically, our approach leverages the facts that (a) there is inherent sensor redundancy when combinations of sensors monitor activities of daily living (ADLs) in smart environments, and (b) the impact of each sensor failure depends on the activities being monitored and the functional redundancy afforded by rest of the heterogeneous sensors available for detecting the activities. Consequently, Idea identifies homes that need to be fixed based on expected degradation in ADL detection performance, and optimizes maintenance scheduling accordingly. We demonstrate that our approach leads to 3--40 times fewer maintenance personnel than a scheme in which failed sensors are fixed without considering their impact.

There is an increasing penetration of smart phones within enterprises. Most smart phone users now run both enterprise as well as personal applications simultaneously on their phones. However, most of the personal apps that are downloaded from public market places are hardly tested for enterprise grade security, and there have been instances of malware appearing in public markets that steal sensitive user information. Smart phone platforms such as Android require users to explicitly provide permissions to applications at install time, yet lack run time monitoring of permission usage by applications. In this paper, we present a framework for the run time enforcement of privacy policies on smart phones, in particular, protecting the privacy of enterprise data on smart phones. Our privacy policies are defined in terms of permissible information flows on the phone during different contexts. This arms users with finer grained control over information access by different applications. In our policy framework, an information flow is defined based on the entities involved in the corresponding inter-process communication(IPC) viz, the caller, callee and the associated IPC data. The information flow policy specifies the conditions under which an IPC flow may be permitted (or denied). Our system tracks information flows at run time and enforces that only flows satisfying all the current policies are permitted on the phone. We describe the design and implementation of our policy based framework in Android, and present performance evaluation results measuring the overhead imposed by our framework.

In this paper, we present a policy based infrastructure for social data access with the goal of enabling scientific research, while preserving privacy. We describe motivating application scenarios that could be enabled with the growing number of user datasets such as social networks and medical datasets. These datasets contain sensitive user information and sufficient caution must be exercised while sharing them with third parties to prevent privacy leaks. One of the goals of our framework is to allow users to control how their data is used, while at the same time enable researchers to use the aggregate data for scientific research. We extend existing access control languages to explicitly model user intent in data sharing as well as supporting additional access modes viz. Complete Access, Abstract Access and Statistical Access that go beyond the traditional allow/deny binary semantics of access control. We then describe our policy infrastructure and show how it can be used to enable the above scenarios while still guaranteeing individual privacy. We then present our initial implementation of the framework extending the SecPAL authorization language to account for new roles and operations.

The age of technology has changed the way that surgeons are being trained. Traditional methodologies for training can include lecturing, shadowing, apprenticing, and developing skills within live clinical situations. Computerized tools which simulate surgical procedures and/or experiences can allow for "virtual" experiences to enhance the traditional training procedures that can dramatically improve upon the older methods. However, such systems do not to adapt to the training context. We describe a ubiquitous computing system that tracks low-level events in the surgical training room (e.g. student locations, lessons completed, learning tasks assigned, and performance metrics) and from these derive the training context. This can be used to create an adaptive training system.