BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//EE - ECPv5.10.0//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:EE
X-ORIGINAL-URL:https://ee.iisc.ac.in
X-WR-CALDESC:Events for EE
BEGIN:VTIMEZONE
TZID:Asia/Kolkata
BEGIN:STANDARD
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
TZNAME:IST
DTSTART:20250101T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=Asia/Kolkata:20250211T150000
DTEND;TZID=Asia/Kolkata:20250211T160000
DTSTAMP:20260526T125901
CREATED:20250206T085840Z
LAST-MODIFIED:20250206T085840Z
UID:241895-1739286000-1739289600@ee.iisc.ac.in
SUMMARY:Talk: Modelling the Switching Dynamics of Advanced Power Semiconductor Devices: From Silicon Superjunction to Wide Bandgap Technologies
DESCRIPTION:Title: Modelling the Switching Dynamics of Advanced Power Semiconductor Devices: From Silicon Superjunction to Wide Bandgap Technologies \nSpeaker: Manish Mandal \nDate: Tuesday\, Feb 11\, 2025 \nTime: 3:00-4:00 pm \nVenue: MMCR \nAbstract: \nThe advancement of power semiconductor devices has significantly transformed modern power conversion systems\, enabling notable enhancements in energy efficiency\, system miniaturization\, and overall performance. Among the emerging technologies\, silicon superjunction MOSFETs (Si SJMOS)\, silicon carbide (SiC) MOSFETs\, and gallium nitride (GaN) high-electron-mobility transistors (HEMTs) have gained prominence in applications such as renewable energy systems\, electric vehicles\, and commercial power supplies. While these devices are commonly available in the 600-650 V range\, SiC MOSFETs extend to higher voltage ratings of 1200-1700 V\, making them well-suited for high-power applications. \nIn power electronic converters\, power semiconductor devices incur switching losses during transitions between their on and off states. Advances in device technology have reduced junction capacitance\, resulting in faster switching transients and lower losses. However\, these improvements also introduce challenges such as oscillations in gate and power loops\, increased electromagnetic interference (EMI)\, crosstalk\, false turn-on events\, and heightened device stress due to the amplified influence of circuit parasitics. Therefore\, an in-depth understanding of switching dynamics is crucial for optimizing device performance and mitigating these issues. \nThis thesis presents a comprehensive investigation into the switching dynamics of advanced power semiconductor technologies (Si SJMOS\, SiC MOSFETs\, and GaN HEMTs). The study employs circuit-based simulations and mathematical modeling to estimate critical performance parameters\, including switching losses\, slew rates of voltage (dv/dt)\, and current (di/dt)\, transition times\, and voltage overshoots. \nThe study begins with developing a mathematical model to characterize the switching transients of Si SJMOS in combination with SiC Schottky barrier diodes (SBDs)\, which mitigate reverse recovery losses. The model employs a nonlinear channel current formulation based on the Nth power law\, effectively capturing the current characteristics in both the ohmic and saturation regions. Additionally\, piecewise nonlinear models are introduced for the gate-drain and drain-source capacitances of Si SJMOS and the reverse-biased capacitance of SiC SBDs. The accuracy of the model is validated using experimental results for three pairs of Si SJMOS and SiC SBD. \nThe investigation then extends to wide bandgap (WBG) devices\, focusing on GaN HEMTs and SiC MOSFETs rated at 600-650 V. A detailed model is developed for GaN HEMTs\, incorporating nonlinear channel current behavior\, junction capacitances\, and parasitic effects. Experimental results for 650 V\, 33 A GaN HEMT validate the accuracy of the model. To represent the switching transients of 650 V SiC MOSFETs\, the existing models originally designed for 1200 V devices are adapted and refined. The model is validated through experimental results for a 650 V\, 30 A SiC MOSFET. \nA comparative analysis is then conducted to evaluate the switching performance of 650 V power semiconductor devices\, including Si SJMOS\, SiC MOSFETs\, and multiple GaN HEMT technologies (e-GaN\, GaN GIT\, and Cascode GaN). Devices with similar voltage (600-650 V) and current (30 A) ratings are assessed in terms of switching losses\, transition times\, (dv/dt\, di/dt)\, and voltage overshoots\, offering valuable insights into device selection for single-phase applications. \nFurther\, the study explores the impact of packaging on the switching behavior of SiC MOSFETs\, particularly in Kelvin-source (TO-247-4) configurations. A detailed model is developed that integrates nonlinear channel current characteristics\, capacitance models\, and circuit parasitic effects. The model is experimentally validated using a 1.2 kV SiC MOSFET. A comparison between TO-247-3 and TO-247-4 packages is also presented\, highlighting the impact of packaging on switching performance. \nIn addition\, an improved model is proposed to predict crosstalk dynamics in SiC MOSFETs. The model incorporates a nonlinear channel current formulation\, parasitic inductances from the package and PCB\, and parasitic capacitances due to PCB layout. These enhancements improve the prediction of (dv/dt)-induced gate-source voltages and the dynamics of false turn-on events. Experimental results for two 1200 V SiC MOSFETs validate the model’s effectiveness. An optimized negative gate voltage and gate resistance design is also proposed to minimize negative gate-source voltage peaks and mitigate false turn-on. \nFinally\, the thesis investigates partial hard turn-on dynamics of SiC MOSFETs in a half-bridge configuration. The study identifies the minimum load current required for zero-voltage switching and quantifies switching losses associated with partial hard turn-on transitions. The findings reveal that these losses deviate significantly from the traditional (1/2)CV2 loss model. Experimental validation is performed using two 1.2kV SiC MOSFETs with varying current ratings.
URL:https://ee.iisc.ac.in/event/talk-modelling-the-switching-dynamics-of-advanced-power-semiconductor-devices-from-silicon-superjunction-to-wide-bandgap-technologies/
LOCATION:MMCR\, Hall C 241\, 1st floor\, EE department
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Kolkata:20250213T153000
DTEND;TZID=Asia/Kolkata:20250213T170000
DTSTAMP:20260526T125901
CREATED:20250212T085203Z
LAST-MODIFIED:20250212T085237Z
UID:241907-1739460600-1739466000@ee.iisc.ac.in
SUMMARY:Title: Tight Frames\, Non-convex Regularizers\, and Quantized Neural Networks for Solving Linear Inverse Problems
DESCRIPTION:Name of the Candidate: Mr. Nareddy Kartheek Kumar Reddy\n\nResearch Supervisor: Prof. Chandra Sekhar Seelamantula\n\nExaminer: Prof. Subhasis Chaudhuri\, EE Dept.\, IIT Bombay\n\nDate and time: February 13\, 2025; 3.30 PM\n\nMeeting Link\n\nTitle: Tight Frames\, Non-convex Regularizers\, and Quantized Neural Networks for Solving Linear Inverse Problems\n \nAbstract:\nThe recovery of a signal/image from compressed measurements involves formulating an optimization problem and solving it using an efficient algorithm. The optimization objective involves data fidelity\, which is responsible for ensuring conformity of the reconstructed signal to the measurement\, and a regularization term to enforce desired priors on  the signal. More recently\, the optimization based solvers have been replaced by deep neural networks.\n\nThis thesis considers three aspects of inverse problems in computational imaging: (i) Choice of data-fidelity term for compressed-sensing image recovery; (ii) Non-convex regularizers in the context of linear inverse problems; and (iii) Explainable deep-unfolded networks and the effect of quantization of model parameters.\n\nPart-1: Tight-Frame-Based Data Fidelity for Compressed Sensing\nThe choice of the sensing matrix is crucial in compressed sensing. Random Gaussian sensing matrices satisfy the restricted isometry property\, which is crucial for solving the sparse recovery problem using convex optimization techniques. However\, tight-frame sensing matrices result in minimum mean-squared-error recovery given oracle knowledge of the support of the sparse vector. If the sensing matrix is not tight\, could one achieve the recovery performance assured by a tight frame by suitably designing the recovery strategy? ­    This is the key question addressed in this part of the thesis.  We consider the analysis-sparse l1-minimization problem with a generalized l2-norm-based data-fidelity and show that it effectively corresponds to using a tight-frame sensing matrix. The new formulation offers improved performance bounds when the number of non-zeros is large. One could develop a tight-frame variant of a known sparse recovery algorithm using the proposed formalism. We solve the analysis-sparse recovery problem in an unconstrained setting using proximal methods. Within the tight-frame sensing framework\, we rescale the gradients of the data-fidelity loss in the iterative updates to further improve the accuracy of analysis-sparse recovery. Experimental results show that the proposed algorithms offer superior analysis-sparse recovery performance. Proceeding further\, we also develop deep-unfolded variants\, with a convolutional neural network as the sparsifying operator. On the application front\, we consider compressed sensing image recovery. Experimental validations on Set11\, BSD68\, Urban100\, and DIV2K datasets show that the proposed techniques outperform the state-of-the-art techniques\, where the performance is measured in terms of peak signal-to-noise ratio (PSNR) and structural similarity index metric (SSIM).\n\nPart 2: Proximal Averaging Methods for Image Restoration and Recovery\nSparse recovery methods are iterative and most techniques typically rely on proximal gradient methods. While the commonly used sparsity promoting penalty is the l1-norm\, which is convex\, alternatives such as the minimax concave penalty (MCP) and smoothly clipped absolute deviation (SCAD) penalty have also been employed to obtain superior results. Combining various penalties to achieve robust sparse recovery is possible\, but the challenge lies in optimal parameter selection. Given the connection between deep networks and unrolling of iterative algorithms\, it is possible to unify the unfolded networks arising from different formulations. We propose an ensemble of proximal networks for sparse recovery\, where the ensemble weights are learnt in a data-driven fashion. The proposed network performs superior to or on par with the individual networks in the ensemble for synthetic data under various noise levels and sparsity conditions. We demonstrate an application to image denoising based on the convolutional sparse coding formulation.\n\nPart 3: Deep Unfolded Networks\, Quantization\, and Explainability\nDeep-unfolded networks (DUNs) have set new performance benchmarks in compressed sensing and image restoration. DUNs are built from conventional iterative algorithms\, where an iteration is transformed into a layer/block of a network with learnable parameters. This work focuses on enhancing the explainability of DUNs by investigating potential reasons behind their superior performance over traditional iterative methods. Our findings reveal that the learned matrices in DUNs are unstable because their singular values exceed unity. However\, the overall DUN gives rise to a recovery accuracy higher than the optimisation techniques. This goes to show that although the linear/affine components of the DUN are unstable\, the overall network is stable\, which leads us to conclude that it is the nonlinearities\, more precisely\, the activation functions\, that are responsible for restoring stability. This study illustrates an intriguing property of deep unfolded networks\, which is not observed in standard optimization schemes.\n\nWe also consider quantization of the network weights for efficient model deployment in resource-constrained devices. Quantization makes neural networks efficient both in terms of memory and computation during inference and also renders them compatible with low-precision hardware deployment. Our learning algorithm is based on a variant of the ADAM optimizer in which the quantizer is part of the forward pass. The gradients of the loss function are evaluated corresponding to the quantized weights while doing a book-keeping of the high-precision weights. We demonstrate applications for compressed image recovery and magnetic resonance image reconstruction. The proposed approach offers superior reconstruction accuracy and quality than state-of-the-art unfolding techniques\, and the performance degradation is minimal even when the weights are subjected to extreme quantization.\n\nImpact of the research: The novel techniques proposed in this thesis led to improved accuracy in linear inverse problems — sparse signals recovery\, compressed image recovery\, image deconvolution\, and image denoising. The tight-frame based algorithms require fewer iterations to converge\, thus reducing the reconstruction time. The quantized neural networks\, on the other hand\, improved the inference time and reduced the model footprint for efficient deployment on the edge. Analysis of deep-unfolded networks has shown that the learnt weights follow a Gaussian distribution suggesting more efficient initialisation schemes than weights derived from ISTA. We also identified potential local instabilities in a deep learning setting\, which are avoided in a conventional optimization setting. The role of the nonlinearity is to restore stability. The analysis showed that while deep unfolded networks have potential instabilities\, they can be useful for solving inverse problems.\n\n\nAbout the Candidate:\nNareddy Kartheek Kumar Reddy is the 13th PhD student to graduate from the Spectrum Lab\, Department of Electrical Engineering at the Indian Institute of Science (IISc). He received a Bachelor of Technology (Honors) degree from Indian Institute of Technology Kharagpur in 2016. Subsequently\, he worked as a Senior Engineer at Honeywell Technology Solutions from 2016 to 2018\, where he focused on developing device drivers for SD card and NAND Flash devices which went into production in Honeywell’s flagship weather radar RDR7000.\n\nKartheek joined IISc as a Masters student in Signal Processing\, and subsequently upgraded to PhD after receiving the prestigious Prime Minister’s Research Fellowship in 2019. He is twice recipient of the Qualcomm Innovation Fellowship\, once in 2020 & again in 2023.
URL:https://ee.iisc.ac.in/event/title-tight-frames-non-convex-regularizers-and-quantized-neural-networks-for-solving-linear-inverse-problems/
LOCATION:Online\, India
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Kolkata:20250217T150000
DTEND;TZID=Asia/Kolkata:20250217T170000
DTSTAMP:20260526T125901
CREATED:20250217T064956Z
LAST-MODIFIED:20250217T065146Z
UID:241948-1739804400-1739811600@ee.iisc.ac.in
SUMMARY:Talk on the four generations of single-neuron models: From the perceptron to the complex adaptive system
DESCRIPTION:Talk on The four generations of single-neuron models: From the perceptron to the complex adaptive system\nby Professor Rishikesh Narayanan\, Molecular Biophysics Unit\, Indian Institute of Science\, Bengaluru 560012\nVenue: Multimedia Classroom (MMCR)\, EE Department\, IISc\nDate & Time: February 17\, 2025\, 3 PM (Coffee will be served at 2.45 PM)\nThe key objective of this talk is to foster interdisciplinary AI research by way of understanding the recent advances in Neuroscience and leveraging them for building superior AI models that are closer to natural intelligence.\n\nAbstract:\nThe first generation of single-neuron models treated neurons as perceptrons or integrate-and-fire devices\, involving some form of summation that was followed by a nonlinearity. This class of models originated in the early 1900s with the law of dynamic polarization laying the conceptual foundation. The 1950s introduced the second generation of models with Hodgkin and Huxley’s ground-breaking use of ordinary differential equations to describe action potential dynamics. This second era emphasized the nonlinear dynamical systems framework to capture ionic interactions underlying neuronal functions. The third era\, beginning in the early 1990s\, incorporated spatial complexity into single-neuron models by acknowledging dendrites as active participants in neural computation. Patch-clamp electrophysiology facilitated discoveries of active conductances in dendrites\, leading to models based on coupled partial differential equations spanning entire dendritic structures. By the early 2000s\, variability among neurons of the same subtype highlighted the need for models beyond a single archetype. This ushered in the fourth generation of models\, where single neurons are recognized as complex adaptive systems. Complex systems are systems where several functionally specialized subsystems interact to yield collective functional outcomes\, and are defined by two key attributes. First\, the interactions among subsystems of a complex system are neither fully determined nor completely random. This intermediate level of randomness is characterized by network motifs — subnetworks that appear more frequently than expected in random networks. The second defining feature of complex systems is degeneracy\, where multiple combinations of distinct subsystems can achieve the same collective function. The complex systems framework unifies earlier models\, highlighting dynamic and adaptive interactions among specialized subsystems to explain collective neuronal function.\n\nAbout Rishi: Rishi earned his Ph.D. from the Department of Electrical Engineering at the Indian Institute of Science\, Bangalore (Advisor: Prof. Y. V. Venkatesh). After that\, he held two postdoctoral positions\, the first at the National Centre for Biological Sciences\, Bangalore (Advisor: Prof. Sumantra Chattarji)\, and the second at the University of Texas at Austin (Advisor: Prof. Daniel Johnston). He returned to the Institute in July 2009. He is currently a Professor at the Molecular Biophysics Unit of the Institute.\n\nHosts: Chandra Sekhar Seelamantula (EE) & Chiranjib Bhattacharyya (CSA)
URL:https://ee.iisc.ac.in/event/talk-on-the-four-generations-of-single-neuron-models-from-the-perceptron-to-the-complex-adaptive-system/
LOCATION:MMCR\, Hall C 241\, 1st floor\, EE department
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Kolkata:20250221T160000
DTEND;TZID=Asia/Kolkata:20250221T173000
DTSTAMP:20260526T125901
CREATED:20250217T054819Z
LAST-MODIFIED:20250221T040549Z
UID:241943-1740153600-1740159000@ee.iisc.ac.in
SUMMARY:EE faculty colloquium
DESCRIPTION:Title: Power Electronics – A Technology Enabler for a Carbon-free Energy Pathway\nSpeaker: Dr. Vinod John\, Dept. of Electrical Engineering\, Indian Institute of Science\nVenue: MMCR\, EE\n            Meeting Link  (for online audience)\nTime: 4 pm\, 21 Feb 2024\n\nAbstract:\nPower electronics technologies play a pivotal role in efficiently transferring electrical energy from sources and storage elements to loads while minimizing power losses. These technologies are critical for optimizing the utilization of renewable energy sources\, enhancing the efficiency of storage systems\, and reducing energy wastage—all of which are essential in the global effort to achieve minimal CO₂ emissions. In this talk\, I will begin with an introduction to the fundamentals of power electronics\, highlighting key research challenges and discussing solutions developed in the Power Electronics Group at the Department of Electrical Engineering\, IISc. Specific focus areas include power converters\, switching devices\, input and output filtering components\, and the control of power conversion systems. I will try to summarize the research efforts over time and indicate a few practical applications of power electronics in real-world scenarios.\n\nSpeaker’s Bio:\nDr. Vinod John is a Professor in the Department of Electrical Engineering at the Indian Institute of Science (IISc)\, Bengaluru. He earned his Ph.D. from the University of Wisconsin–Madison and subsequently worked at GE Research\, New York\, and Northern Power Systems\, Vermont\, before joining IISc. His research interests encompass\npower electronics\, switched-mode power conversion\, distributed energy resources\, and energy storage systems.
URL:https://ee.iisc.ac.in/event/ee-faculty-colloquium-2/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Kolkata:20250227T140000
DTEND;TZID=Asia/Kolkata:20250227T160000
DTSTAMP:20260526T125901
CREATED:20250224T093957Z
LAST-MODIFIED:20250224T093957Z
UID:241962-1740664800-1740672000@ee.iisc.ac.in
SUMMARY:EE  PhD  Defense: Infimal Convolution Based Regularization   for Image recovery
DESCRIPTION:Student : Deepak G Skariah \nAdvisor : Prof. Muthuvel Arigovindan \nTitle :  Infimal Convolution Based Regularization   for Image recovery \nDate and Time:   27.02.2025 (Thursday)\,  2 pm. \nVenue :  MMCR\, Department of Electrical Engineering \n Meeting link \nThesis examiners:   Prof.  Kedar Khare\,  Prof. Naren Nayak \nDefense examiner:   Prof.  Kedar Khare \nAbstract\nThe quality of image captured by acquisition devices has increased drastically over the years largely due to a revolution in imaging sensor capability. But\, image acquisition under low illumination continues to be a bottleneck for imaging devices such as  optical microscopes   leading to blurred and noisy images.  A potential solution to this limitation   is a computational approach known as image restoration. An image restoration   algorithm recovers  an estimate of the original image from a noisy blurred observation  while assuming a knowledge of the image degradation model.  The restoration problem is even more challenging when it comes to a spatio-temporal signal as a good restoration scheme needs to be mindful of presence of motion in the measured signal. This means that in spatio-temporal signal restoration problem\, the algorithm should ensure temporal regularity of restored signal in addition to spatial regularity. Regularization based image restoration attempts to pose image restoration problem as a regularized optimization problem from the measured signal.  We propose to exploit the concept of infimal convolution from convex analysis to design effective and efficient restoration schemes for images and spatio-temporal images. \nIn our first work\, we address the problem of regularization design. We   propose  a family of derivative based regularization which we call generalized unitary invariant regularization and it belongs to class of infimal convolution based functionals. We  also design an algorithmic scheme to optimize the resultant optimization problem. We demonstrate the quality of proposed algorithm and restoration scheme through multiple experiments on simulated data. \nIn our  second work\, we address the restoration of spatio-temporal images measured from TIRF microscopes where a sequence of noisy blurred images are observed over time. We once again exploit the infimal convolution based approach to design a novel spatio-temporal regularizer that is tailor made for above class of signals. The proposed regularization was designed to ensure both  spatial and temporal regularity of restored signal. The resultant regularization functional is defined as an optimization problem where the cost is a weighted sum of two constituent functions where the two functions play the role of promoting spatial and temporal regularity respectively.   We also design an algorithm to optimize the resultant restoration problem using this regularization. We demonstrate the quality of the proposed algorithm by testing the restoration quality against spatio-temporal measurements    collected from TIRF microscopes. \nIn the third and final work we address the problem of estimating the relative weights in spatio-temporal regularization functional designed based on infimal convolution formulation. We propose a renewed optimization model where the spatio-temporal signal is estimated together with the better quality image estimate by incorporating the weights as part  of the optimization problem. We also design an iterative scheme to optimize the resultant joint optimization model. We demonstrate the effectiveness of this scheme against other  joint optimization schemes for spatio-temporal signal estimation.
URL:https://ee.iisc.ac.in/event/ee-phd-defense-infimal-convolution-based-regularization-for-image-recovery/
LOCATION:Multi-Media Class Room (MMCR)\, EE Department (Hybrid mode)
END:VEVENT
END:VCALENDAR