BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.2.3.1//EN
TZID:Asia/Kolkata
X-WR-TIMEZONE:Asia/Kolkata
BEGIN:VEVENT
UID:51@cds.iisc.ac.in
DTSTART;TZID=Asia/Kolkata:20240425T100000
DTEND;TZID=Asia/Kolkata:20240425T110000
DTSTAMP:20240417T072125Z
URL:https://cds.iisc.ac.in/events/ph-d-thesis-colloquium-cds-seminar-hall-
 102-learning-deep-neural-networks-from-limited-and-imperfect-data/
SUMMARY:Ph.D. Thesis {Colloquium}: CDS Seminar Hall # 102 "Learning Deep Ne
 ural Networks From Limited and Imperfect Data"
DESCRIPTION:DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES\nPh.D. Thesis Col
 loquium\n\n\n\nSpeaker : Mr. Harsh Rangwani\n\nS.R. Number : 06-18-01-10-1
 2-19-1-17477\n\nTitle :"Learning Deep Neural Networks From Limited and Imp
 erfect Data"\nResearch Supervisor: Prof. Venkatesh Babu\nDate &amp\; Time 
 : April 25\, 2024 (Thursday) at 10:00 AM\nVenue : # 102 CDS Seminar Hall\n
 \n\n\nABSTRACT\n\nDeep Neural Networks have demonstrated orders of magnitu
 de improvement in capabilities over the years after AlexNet won the ImageN
 et challenge in 2012. One of the major reasons for this success is the ava
 ilability of large-scale\, well-curated datasets. These datasets (e.g.\, I
 mageNet\, MSCOCO\, etc.) are often manually balanced across categories (cl
 asses) to facilitate learning of all the categories. This curation process
  is often expensive and requires throwing away precious annotated data to 
 balance the frequency across classes. This is because the distribution of 
 data in the world (e.g.\, internet\, etc.) significantly differs from the 
 well-curated datasets and is often over-populated with samples from common
  categories. The algorithms designed for well- curated datasets perform su
 boptimally when used to learn from imperfect datasets with long-tailed imb
 alances and distribution shifts. For deep models to be widely used\, getti
 ng away with the costly curation process by developing robust algorithms t
 hat can learn from real-world data distribution is necessary. Toward this 
 goal\, we develop practical algorithms for Deep Neural Networks that can l
 earn from limited and imperfect data present in the real world. This thesi
 s is divided into four segments\, each covering a scenario of learning fro
 m limited or imperfect data. The first part of the thesis focuses on Learn
 ing Generative Models for Long-Tail Data\, where we mitigate the mode-coll
 apse for tail (minority) classes and enable diverse aesthetic image genera
 tions as head (majority) classes. In the second part\, we enable effective
  generalization on tail classes through Inductive Regularization schemes\,
  which allow tail classes to generalize as the head classes without enforc
 ing explicit generation of images. In the third part\, we develop algorith
 ms for Optimizing Relevant Metrics compared to the average accuracy for le
 arning from long-tailed data with limited annotation (semi-supervised)\, f
 ollowed by the fourth part\, which focuses on the effective domain adaptat
 ion of the model to various domains with zero to very few labeled samples.
 \n\nGenerative Models for Long-Tail Data. We first evaluate generative mod
 els’ performance\, specifically variants of Generative Adversarial Netwo
 rks (GANs) on long-tailed datasets. The GAN variants suffer from either mo
 de-collapse or miss-class modes during generation. To mitigate this\, we p
 ropose Class Balancing GAN with a Classifier in the Loop\, which uses a cl
 assifier to asses the modes in generated images and regularizes GAN to pro
 duce all classes equally. To alleviate the dependence on the classifier\, 
 following our observation that spectral norm explosion of Batch Norm param
 eters is the major reason for mode collapse. We develop an inexpensive gro
 up Spectral Regularizer (gSR) to mitigate the spectral collapse\, signific
 antly improving the SotA conditional GANs (SNGAN and BigGAN) performance o
 n long-tailed data. However\, we observed that class confusion was present
  in the generated images due to norm regularization. In our latest work\, 
 NoisyTwins\, we factor the latent space as distinct Gaussian by design for
  each class\, enforcing class consistency and intra-class diversity using 
 a contrastive approach (BarlowTwins). This helps to scale high-resolution 
 StyleGANs for ≥ 1000 class long-tailed datasets of ImageNet-LT and iNatu
 ralist2019\, achieving state-of-the-art (SotA) performance.\n\nInducting R
 egularization Schemes for Long-Tailed Data. While Data Generation is excit
 ing for improving classification models on tail classes\, it often comes w
 ith the cost of training an auxiliary GAN model. Hence\, a lightweight tec
 hnique like enhancing loss weights (re-weighting) for tail classes while t
 raining CNNs is practical to improve minority class performance. However\,
  despite this\, the model only attains minima for the head class loss and 
 converges to saddle point for tail classes. We show that inducing inductiv
 e bias of escaping saddles and converging to minima for tail classes\, usi
 ng Sharpness Aware Minimization (SAM) significantly improves performance o
 n tail classes. Further training Vision Transformer (ViT) for long-tail re
 cognition is hard\, as they don’t have inductive biases like locality of
  features\, which makes them data hungry. We propose DeiT-LT\, which intro
 duces OOD and low-rank distillation from CNN to induce CNN-like robustness
  into scalable ViTs for robust performance.\n\nSemi-Supervised Long-Tailed
  Learning. The above methods work in supervised long-tail learning\, where
  they avoid throwing off the annotated data. However\, the real benefit of
  long-tailed methods could be leveraged when they utilize the extensive un
 labeled data present (i.e.\, semi-supervised setting). For this\, we intro
 duce a paradigm where we measure the performance using relevant metrics li
 ke worst-case recall and recall H-mean on a held-out set\, and we use thei
 r feedback to learn in a semi-supervised long-tailed setting. We introduce
  Cost-Sensitive Self Training (CSST) generalizes self-training (e.g.\, Fix
 Match\, etc.) based semi-supervised learning to long-tail settings with st
 rong guarantees and empirical performance. The general trend these days is
  to use self-supervised pre-training to obtain a robust model and then fin
 e-tune it. In this setup\, we introduce SelMix\, an inexpensive fine-tunin
 g technique to optimize the relevant metrics using pre-trained models. In 
 SelMix\, we relax the assumption that unlabeled distribution is similar to
  the labeled\, making models robust to distribution shifts.\n\nEfficient D
 omain Adaptation. The long-tail learning algorithms focus on limited data 
 setup and improving in-distribution generalization. Still\, for practical 
 usage\, the model must learn from imperfect data and perform well across v
 arious domains. Toward this goal\, we develop Submodular Subset Selection 
 for Adversarial Domain Adaptation\, which carefully selects a few samples 
 to be labeled for maximally improving model performance in the target doma
 in. To further improve the efficiency of the Adaptation procedure\, we int
 roduce Smooth Domain Adversarial Training (SDAT)\, which converges to gene
 ralizable smooth minima. The smooth minimum enables efficient and effectiv
 e model adaptation across domains and tasks.\n\n\n\nALL ARE WELCOME
CATEGORIES:Events,Ph.D. Thesis Colloquium
END:VEVENT
BEGIN:VTIMEZONE
TZID:Asia/Kolkata
X-LIC-LOCATION:Asia/Kolkata
BEGIN:STANDARD
DTSTART:20230426T100000
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
TZNAME:IST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR