NCC 2025
2025 national conference on communications
6th-9th March, 2025
Indian Institute of Technology Delhi
TUTORIALS
Date: 6th March
Morning Half Sessions (9Am-1PM)
Title: Optimization Algorithms for Distributed Machine Learning
Speaker: Bharath B N, IIT Dharwad
Venue: LH212
Abstract: In recent years, with rapid improvement in technologies, there is an enormous amount of distributed data generated at various devices such as sensors, smartphones, healthcare devices, and computers. In traditional Machine Learning (ML), the data generated from various devices are usually collected and stored centrally for further processing and optimization tasks. However, collecting the data centrally is not always feasible as this gives rise to privacy concerns. Moreover, sending the raw data to the central server incurs huge communication costs. This has made distributed learning, which requires distributed optimization, a necessary choice for learning good ML models.
In this tutorial, we provide the necessary background by first understanding various algorithms for single node setting, also sometimes referred to as centralized setting. In particular, we begin by studying convergence properties of classical algorithms such as Gradient Descent (GD) and Stochastic GD (SGD) under both convex and non-convex settings. Next, we discuss the disadvantages of the SGD such as high variance, and provide remedies to reduce the variance. Several variance reduction techniques have been proposed in the literature. In this tutorial, we focus on one variance reduction technique called SAGA and provide pointers to other variance reduction techniques. Although SGD and its variants are very well studied in the single node case, the decentralized version of these algorithms poses added challenges such as model drift, communication delays etc. Hence, there is a need to make the decentralized versions of SGD faster.
The background on the centralized setting provides necessary tools to study the distributed learning algorithms. A de-facto method to solve the distributed learning problem is called Federated Averaging (FedAvg) algorithm. We first provide the proof of convergence of the classical FedAvg, and underline its disadvantages, including model drift, impact of heterogeneity in the data and divergence issues under relaxed assumptions. Several algorithms have been proposed to tackle the disadvantages of the FedAvg algorithm. Towards the end of the tutorial, we will study these algorithms which include SVRG, FedProx, SCAFFOLD and many others. In summary, we will provide a theoretical understanding of distributed learning problems. The tutorial is designed to enable research students and industry professionals to gain enough background to start working in the exciting area of distributed learning.
Bio: Dr. Bharath B. N. is currently working as an assistant professor in the Department of electrical, electronics and communication engineering at IIT Dharwad. He received his B. E. in electrical and electronics engineering from B. M. S. college of engineering in Bengaluru in 2005. He received his (direct) Phd from the electrical communication engineering department at IISc in 2013. He worked as a senior engineer at Qualcomm Inc, Bengaluru from 2013 to 2014. From 2014 to 2017, he worked as a faculty in the department of electronics and communication engineering at PES, Bangalore south campus, Bengaluru. Since 2017, he has been with IIT Dharwad. His research interests include federated learning, bilevel optimization, online optimization/learning, distributed optimization and signal processing for communication problems.
Title: Building 3GPP Compliant Cell-Free Networks: Breaking Barriers, Enhancing Connectivity
Speakers:
Abhay Kumar Sah, IIT Roorkee
Adarsh Patel, IIT Mandi
Venue: LH213
Abstract: To meet the ever-growing demand for higher data rates, improved reliability, and energy efficiency, cell-free networks have emerged as a promising alternative to conventional cellular architectures. These networks distribute antennas across a large geographical area via multiple access points (APs), ensuring uniform data rates for users. Unlike traditional cellular systems, where users are surrounded by a base station, cell-free networks enable a scenario where users are surrounded by multiple APs, all connected to a central processing unit through a backhaul link, fostering seamless and ubiquitous connectivity.
This tutorial aims to provide the theoretical foundations, architectural design, and practical implementation challenges of 3GPP-compliant cell-free massive MIMO (CF-mMIMO) networks, being developed at IIT Roorkee and IIT Mandi in collaboration with MathWorks Inc. and C-DOT. The session will trace the evolution from multi-cell to cell-free massive MIMO, highlighting key advancements in the 5G physical layer and function splits for centralized and distributed CF-mMIMO realizations. Attendees will gain insights into the design and development of a CF-mMIMO simulator comparing its performance with traditional cellular networks, which leverages sounding signal for uplink channel estimation and downlink precoder design. The tutorial is tailored for researchers, engineers, and industry professionals seeking a deeper understanding of next-generation wireless communication networks, aligning with the 3GPP framework and future 6G aspirations.
Bios:
Abhay Kumar Sah: Abhay Kumar Sah received his B.Tech. degree in Electronics and Communications Engineering from Jaypee Institute of Engineering & Technology, Guna, in 2010, and his Ph.D. degree from the Department of Electrical Engineering at the Indian Institute of Technology Kanpur in 2017. Following his Ph.D., Dr. Sah joined Samsung R&D Institute Bangalore as a Senior Lead Engineer, where he played a key role in developing 3GPP Release 14 features. He later served as a Lead Software Architect at Radisys India Pvt. Limited, Bangalore, where he was instrumental in the development of PUSCH algorithms for 5G base stations. Currently, he serves as an Assistant Professor in the Department of Electronics & Communication Engineering at Indian Institute of Technology Roorkee. His research interests focus on realizing cell-free massive MIMO systems for next-generation wireless networks and exploring the applications of deep learning in wireless communication systems. He has authored over 30 research articles and holds more than 15 global and Indian patents. Dr. Sah has been recognized with several prestigious awards, including the MathWorks Research Award, DoT’s Telecom Technology Development Fund, SERB Start-up Research Grant, DST-Inspire Faculty Award, TCS Research Fellowship, and the Vice-Chancellor’s Gold Medal for outstanding academic performance during his bachelor’s degree. He also received the Samsung Citizen Award for his contributions to 3GPP Release 14 features.
Adarsh Patel: Adarsh Patel is an assistant professor in the School of Computing and Electrical Engineering (SCEE) at Indian Institute of Technology Mandi, Himachal Pradesh, India. His research group hail from the InfoComm lab at IIT Mandi. His research interests are in the areas of the next-generation Wireless Networks, which includes massive MIMO and mm-Wave, IoT, Integrated sensing and communications (ISC), Cognitive Radio and License Assisted Access, Cooperative Communication, Molecular Communications, Sensor Networks, and others having applications of Signal Processing, Game Theory, Machine/ Deep Learning, Tensors, Optimization Techniques, etc. He is currently heading the InfoComm Lab and the 5G use case lab at IIT Mandi. He is an active member of the Telecommunications Standards Development Society (TSDSI) and associated working groups, International Telecommunication Union (ITU), Indian National Academy of Engineering (INAE), Institute of Electrical and Electronics Engineers (IEEE). He received a Ph.D. degree in Electrical Engineering (EE) from the Dept. of EE, IIT Kanpur, in 2017. His PhD thesis work received the Innovative Student Project Award 2017 by the Indian National Academy of Engineering (INAE). He has worked as a postdoctoral research staff member in the Sensor Fusion Lab in the Department of EE & CSE at Syracuse University, NY, USA. He is a recipient of the DoT-Telecom Technology Development Fund and SERB-Startup Research Grant, TCS Research fellowship, SERB-National Postdoctoral fellowship to conduct research at IIT Mandi, IIT Kanpur, and IISc Bangalore, respectively.
AFTERNOON HALF SESSIONS (2PM-5PM)
Title: STARS: Unlocking the Potential of 6G with Simultaneously Transmitting and Reflecting Surfaces
Speakers:
Sarvendranath Rimalapudi, IIT Tirupati
Salil Kashyap, IIT Guwahati
Venue: LH212
Abstract: Simultaneously transmitting and reflecting surface (STARS) is one of the recent innovations, which is envisioned to unlock the potential of 6G by enhancing wireless network capacity and energy efficiency to meet the ever-increasing demand for higher data rates. It can simultaneously transmit and reflect the incident electromagnetic signal, enabling 360 degree coverage unlike the traditional intelligent reflecting surfaces (IRS), which can only reflect the incident signal. The circuit impedance of each STARS element can be configured to control the amplitude and phase shifts induced to the transmitted and reflected signals. Therefore, STARS can be configured to improve the desired signal strength or to reduce interference at a specific location in all directions, which enables us to build smart wireless networks supporting a wide range of applications.
The tutorial first introduces the concept of reflective surfaces and signal model without assuming any prior knowledge. We will discuss testbeds built with different reflective surfaces, showcasing their practical implementation feasibility. We will highlight the key differentiating features of the STARS and explain its practical use cases. Following that, the model for the transmitted and reflected signals will be introduced to highlight the impact of transmit and reflect amplitude and phase shift configurations. We then discuss the different phase configuration strategies and the three main operating modes, namely, energy splitting, mode switching, and time switching. This will be followed by an introduction to the training schemes for estimating channel state information that is required to implement these phase shift configuration strategies.
The tutorial will then shift to a detailed analysis of two critical aspects: diversity and power scaling. We will look at how these factors affect the system along with strategies for balancing power allocation between transmitting and reflecting users. We will then discuss different multiple antenna techniques combined with transmit and reflect beamforming at the STARS to improve the overall system performance. We will introduce different problem formulations for orthogonal and non-orthogonal multiple access systems and unicast and multicast scenarios considering different operating modes. Finally, the tutorial will conclude by introducing generalizations to reflective surfaces and with a detailed discussion of future research directions and open problems.
The tutorial has been structured in a classroom teaching format with a focus on fundamentals and a balance between theoretical concepts and practical applications. It aims to provide valuable insights and a quick start to the research scholars and industry professionals into the exciting field of STARS, which can be a key enabler for 6G.
Bios:
Sarvendranath Rimalapudi: Sarvendranath Rimalapudi received the B.Tech. degree in Electrical and Electronics Engineering from the National Institute of Technology Karnataka, Surathkal, India, in 2009, and the M.E. and Ph.D. degrees from the Department of Electrical Communication Engineering, Indian Institute of Science (IISc), Bangalore, India, in 2012 and 2020, respectively. He is currently an Assistant Professor at the Department of Electrical Engineering, Indian Institute of Technology (IIT) Tirupati, Tirupati, India. In 2021, he was a Postdoctoral Researcher with the Department of Electrical Engineering, Linköping University, Linköping, Sweden. He also served as an Assistant Professor with the Department of Electronics and Electrical Engineering, IIT Guwahati, from January 2022 to June 2023. From 2012 to 2016, he was with Broadcom Communications Technologies, Bangalore, where he worked on developing and implementing algorithms for LTE and IEEE 802.11ac wireless standards. Dr. Sarvendranth was awarded the Kaikini Research Fellowship in 2017, which was granted to one student among all engineering departments at IISc. He received the INSPIRE Faculty Fellowship in 2022. His research interests include machine learning for wireless communication, multiple antenna techniques, intelligent reflecting surfaces, spectrum sharing, and next-generation wireless standards.
Salil Kashyap: Salil Kashyap is an Associate Professor in the Dept. of Electronics and Electrical Eng. at IIT Guwahati. Before joining IIT Guwahati, he was a senior DSP Engineer at Marvell where he designed physical layer algorithms for next generation WLANs (IEEE 802.11ax). Prior to that, he was a post-doctoral researcher at Linköping University, Sweden. He received his PhD from IISc Bangalore, M.Tech from IIT Guwahati and B.Tech from NERIST. His research spans areas of wireless communications and signal processing with emphasis on mathematical modeling, performance analysis and algorithm design for 5G and beyond 5G communication systems.
Title: Uncertainty Estimation for Trustworthy AI
Speakers:
Vipul Arora, IIT Kanpur
Parampreet Singh, IIT Kanpur
Venue: LH213
Abstract: Deep learning models have achieved remarkable success across various modalities, including images, audio, and text. However, their tendency to produce overconfident predictions, even when incorrect, poses a significant challenge for deployment in safety-critical applications such as autonomous driving and disease diagnosis.
Uncertainty estimation addresses this issue by enabling deep learning models to assess the reliability of their predictions. It provides a mechanism for determining whether a model’s output should be trusted, thereby allowing for informed decision making.
Recent research in Uncertainty estimation has focused primarily on two tasks: confidence calibration and out-of-distribution (OOD) detection. Confidence calibration techniques aim to align a model’s predicted probabilities with its actual accuracy, mitigating overconfidence issues. OOD detection, on the other hand, helps identify inputs that fall outside the distribution of the training data, ensuring that the model does not produce misleadingly confident predictions for unfamiliar examples. Several approaches have been proposed to address these challenges, including Bayesian methods, Monte Carlo dropout, deep ensembles, and post-hoc calibration techniques.
This tutorial will provide a comprehensive overview of uncertainty estimation, covering key methodologies, practical applications, and evaluation metrics. There will be specific examples of audio applications benefitting from uncertainty estimation. Attendees will gain insights into the strengths and limitations of existing techniques and learn best practices for integrating uncertainty-aware models into real-world systems.
Bios:
Vipul Arora: Vipul works on developing learning based methods mostly for audio processing (music, speech, and other sounds). Vipul has interest in Physics, and so he works with Physicists to develop learning based methods for problems in computational Physics too.
Vipul's research contributes to (i) audio representation learning for tasks such as audio search, audio event detection and audio annotation, (ii) human-machine learning for accelerating data annotation and for trustworthy machine learning with application to speech, music and audio analysis, (iii) generative machine learning for enhancing Monte Carlo simulations for studying lattices in statistical and particle Physics, and (iv) developing advanced learning-based tools to enhance air quality monitoring at scale.
Vipul received his B.Tech. and Ph.D. degrees in Electrical Engineering from the Indian Institute of Technology (IIT) Kanpur, India. Vipul's Ph.D. thesis was titled “Analysis of Pitched Polyphonic Music for Source Transcription”, where he worked on analyzing music audio to identify and transcribe different instruments/voices playing simultaneously. During postdoc at Oxford University (UK), he developed speech recognition systems using linguistic principles, with applications in automatic language teacher and speech recognition for low-resource languages. At Amazon in Boston (USA), he worked on audio classification for developing Alexa home security system, with research focusing on classification with imbalanced data.
Parampreet Singh: Parampreet Singh received his B.Tech. from Guru Nanak Dev Engineering College, Ludhiana, in 2019, and an M.Tech. from the National Institute of Technology, Jalandhar, in 2021. He is currently a Ph.D. Research Scholar at the Indian Institute of Technology, Kanpur, under the supervision of Prof. Vipul Arora.
His research focuses on Machine Learning and Signal Processing for audio and music applications, with a particular emphasis on self-supervised learning, uncertainty estimation, and Explainable AI. As a trained musician with a decade of experience in Gurmat Sangeet, a variant of Hindustani classical music from Punjab, he integrates domain knowledge with AI-driven methodologies and mathematical analysis to advance computational music and audio analysis.