BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.4.0.1//EN
TZID:Asia/Kolkata
X-WR-TIMEZONE:Asia/Kolkata
BEGIN:VEVENT
UID:219@cds.iisc.ac.in
DTSTART;TZID=Asia/Kolkata:20260807T120000
DTEND;TZID=Asia/Kolkata:20260807T130000
DTSTAMP:20260803T144636Z
URL:https://cds.iisc.ac.in/events/ph-d-thesis-colloquium-102-cds-07-august
 -2026systems-for-federated-learning-across-the-edge-fog-cloud-continuum/
SUMMARY:Ph.D: Thesis Colloquium: 102: CDS: 07\, August 2026“Systems for F
 ederated Learning across the Edge–Fog–Cloud Continuum”
DESCRIPTION:DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES\nPh.D. Thesis Col
 loquium\n\n\n\nSpeaker : Ms. Roopkatha Banerjee\nS.R. Number : 06-18-01-10
 -12-21-1-19488\nTitle : “Systems for Federated Learning across the Edge
 –Fog–Cloud Continuum”\nResearch Supervisor : Prof. Yogesh Simmhan\nD
 ate &amp\; Time : August 07\, 2026 (Friday)\, 12:00 Noon\nVenue : #102\, C
 DS Seminar Hall\n\n\n\nABSTRACT\n\n“Federated Learning (FL) has emerged 
 as a privacy-preserving paradigm for training deep neural networks(DNNs) a
 cross distributed clients that own private data\, sharing only model updat
 es rather than raw data with a coordinating server. Its promise is amplifi
 ed by accelerated edge devices\, such as NVIDIA Jetson\, which pack hundre
 ds to thousands of CUDA cores into compact\, low-power form factors co-loc
 ated with data sources in smart cities\, homes\, and industrial settings. 
 Yet most FL research is validated in pseudo-distributed simulation\, under
  idealized assumptions: homogeneous and reliable clients\, unconstrained e
 nergy\, trusted infrastructure\, and labeled\, stationary data from a fixe
 d set of classes. These assumptions rarely hold when FL is deployed on rea
 l\, heterogeneous edge and fog hardware in the field.\n\nThis dissertation
  takes a systems approach to making FL practical under the constraints tha
 t real-world edge deployments impose. Motivated throughout by smart-city d
 eployments\, we build a modular framework as a common substrate and layer 
 upon it optimizations and scheduling strategies that each confront a const
 raint simulation hides: device heterogeneity and failure\, energy budgets\
 , continuous drift and data sovereignty\, and unlabeled\, open-world data.
  All are validated on real edge accelerators and single-board computers sp
 anning the edge–fog–cloud continuum\, at the scale of hundreds of clie
 nts.\n\nWe first develop Flotilla\, a scalable\, modular\, and resilient F
 L framework that serves as the deployment substrate for this dissertation.
  Its "user-first" design lets researchers rapidly compose synchronous and 
 asynchronous FL strategies while remaining agnostic to the DNN architectur
 e\, and its stateless clients and checkpointed server state enable rapid r
 ecovery from failures. Across five FL strategies and five DNN models\, Flo
 tilla shows sub-second failover on 200+ clients and a resource footprint c
 omparable to or better than Flower\, OpenFL\, and FedML.\n\nBuilding on th
 is substrate\, FedJoule trains within a global energy budget over heteroge
 neous edge accelerators. We formulate a client-selection problem that maxi
 mizes accuracy within an overall energy limit while reducing training time
 \, solved through a bi-level Integer Linear Program using approximate Shap
 ley values and energy–time prediction models. Across diverse budgets and
  non-IID distributions\, FedJoule outperforms state-of-the-art and simple 
 baselines by 15% on accuracy and 48% on time.\n\nWe then design SURGE\, a 
 continuous FL framework that remains dependable and data-sovereign under r
 eal-world drift. SURGE decouples data storage from computation using decen
 tralized W3C Solid pods\, and delegates local training to trusted\, GPU-eq
 uipped fog nodes through drift-aware pod selection\, trust-constrained pod
 -to-fog assignment\, and pipelined scheduling. Across deployments of 20–
 160 devices\, SURGE improves global-model accuracy by 3–17% over FedDrif
 t and reduces drift-detection overhead by up to 23x\, adding only sub-seco
 nd Solid overhead and roughly 7% orchestration overhead for Flotilla.\n\nF
 inally\, in ODIN\, we support incremental FL and open-world class discover
 y over unlabeled\, continuously evolving data streams. ODIN maps data into
  a hyperspherical feature space using a lightweight Vision Transformer\, f
 ilters known from out-of-distribution samples\, and coordinates a server-s
 ide merge-and-discover protocol to detect novel categories\, while diversi
 ty-based coreset replay mitigates catastrophic forgetting. On two real-wor
 ld Indian vehicle datasets\, ODIN outperforms the FC²DL and Fed-GCD basel
 ines by 7–32% and 19–41%\, respectively\, approaching SAM-based zero-s
 hot labeling to within 12% at orders-of-magnitude lower latency.\n\nTogeth
 er\, these contributions offer a systems foundation for federated learning
  that is deployable\, efficient\, dependable\, and adaptive under real-wor
 ld edge constraints\, giving practitioners concrete frameworks\, optimizat
 ion methods\, and scheduling strategies to leverage accelerated edge and f
 og hardware for privacy-preserving distributed intelligence.”\n\n\n\n AL
 L ARE WELCOME
CATEGORIES:Events,Ph.D. Thesis Colloquium
END:VEVENT
BEGIN:VTIMEZONE
TZID:Asia/Kolkata
X-LIC-LOCATION:Asia/Kolkata
BEGIN:STANDARD
DTSTART:20250807T120000
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
TZNAME:IST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR