Synapse

 

Monthly Newsletter from CNI

Issue - December 2024

Discovering Research Opportunities: A Visit to  CNI by TTIT students

Students from Dr. T. Thimmaiah Institute of Technology recently visited CNI for a brief tour focused on research and internship opportunities. During their visit, they interacted with CNI scholars, gaining insights into ongoing research projects. The discussions highlighted various internship programs available at CNI, which can provide valuable hands-on experience for students. Additionally, the students toured the CNI laboratories, where they learned about the architecture and functioning of the servers used in research. This visit offered a practical perspective on the academic environment at CNI and helped students understand the importance of research in their fields. Overall, the experience was beneficial in connecting students with potential future opportunities.

Differentially Private Release of Spatio-Temporal Data Statistics 

In this talk, Dr. V. Aravind Rameshwar discussed his work at the Indian Urban Data Exchange (IUDX). In his talk, he introduced differential privacy (DP), the sample mean, and the variance of a given dataset and described the notion of user-level DP. He introduced the techniques to reconstruct the statistics, the worst-case error involved during the reconstruction, and its importance in improving user privacy. After a thorough introduction, he explained the problem statement, and using a classical bound, he described the proposed novel algorithm that enhances privacy loss. He tested this algorithm on a real-world Intelligent Traffic Management System (ITMS) dataset collected from an Indian city. He showed that the privacy loss incurred by his algorithm is comparatively lesser than in the existing literature.    

Need for Speed: Minimizing the impact of Bufferbloat

In this talk, Prof. Mohit Tahiliani touched upon the technological advancements in networking services, both in software and hardware. Then, he explained a problem buffer bloat and its impact on the specific latency applications. He explained the existing queue disciplines, such as CoDel and PIE, which were designed to reduce buffer bloat. He described a bug observed during the PIE implementation in the Linux kernel and their quick fix. Furthermore, he mentioned that the Linux community approved this quick fix.

Design and Analysis of Low Complexity Techniques for IRS-Aided Wireless Communications

In this presentation, Yashvanth L explored low-complexity techniques for the design of intelligent reflecting surfaces (IRS), in wireless communication systems. He has begun by examining IRS optimization through random configuration tuning and opportunistic user scheduling, demonstrating that multi-user diversity can achieve optimal IRS performance without explicit optimization in systems with many users. He addressed the impact of an IRS, as a passive device without bandpass filters, on the performance of other mobile network operators when controlled by a single operator. This analysis considered both sub-6 GHz and millimeter wave (mmWave) frequency bands in centralized and distributed deployment scenarios. Finally, he tackled challenges in wideband beamforming with IRS technology, focusing on beam-split effects that can degrade array gain and throughput. He proposed two low-complexity strategies: a distributed IRS approach to mitigate beam-split effects and an opportunistic OFDMA method to exploit them positively. The theoretical findings will be supported by numerical experiments.

Interventional Causal Representation Learning

In this talk, Prof. Tajer Ali, explained about Causal representation learning (CRL). He described that, CRL aims to develop a causal understanding of the world by learning appropriate representations that facilitate causal interventions, reasoning, and planning.  Specifically, CRL involves a data-generating process in which high-level, causally related latent variables are mapped to low-level, typically high-dimensional observed data through a transformation. The CRL process utilizes observed data to recover the causal structure, latent causal variables, and the unknown transformation. He provided a comprehensive overview of the problem space and recent advancements in various aspects of interventional CRL, focusing on both identifiability (information-theoretic) and achievability (algorithmic) dimensions.

Efficient Repair of Reed-Solomon Codes and Tamo-Barg Codes

In this talk, Prof. Lalitha Vadlamani delivered a talk on the efficient repair of Reed-Solomon codes, which are polynomial evaluation codes that can be effectively repaired by treating code symbols as vectors over a subfield. She introduced the trace-repair framework developed by Guruswami-Wootters, enabling efficient repair of these codes. Additionally, she presented an optimal construction of Reed-Solomon codes by Tamo et al., which achieves the cut-set bound.

The discussion also covered Tamo-Barg codes, a class of optimal locally repairable codes (LRCs) that utilize Reed-Solomon codes as their local codes. In scenarios involving single node failures, repair occurs solely within local groups. The speaker highlighted that the repair bandwidth can be minimized using Reed-Solomon repair techniques. Furthermore, she provided a construction of Tamo-Barg codes where the local Reed-Solomon codes can be optimally repaired and established connections between this class of codes and those with local regeneration capabilities.