Seminars and Talks

Are Large-scale Datasets Necessary for Self-Supervised Pre-training?
by Alaa El-Nouby
Date: Friday, Mar. 11
Time: 15:30
Location: Online Call via Zoom

Our guest speaker is Alaa El-Nouby from Meta AI Research and Inria Paris and you are all cordially invited to the CVG Seminar on March 11th at 3:30 p.m. on Zoom (passcode is 913674).

Abstract

Pre-training models on large scale datasets, like ImageNet, is a standard practice in computer vision. This paradigm is especially effective for tasks with small training sets, for which high-capacity models tend to overfit. In this work, we consider a self-supervised pre-training scenario that only leverages the target task data. We consider datasets, like Stanford Cars, Sketch or COCO, which are order(s) of magnitude smaller than Imagenet. Our study shows that denoising autoencoders, such as BEiT or a variant that we introduce in this paper, are more robust to the type and size of the pre-training data than popular self-supervised methods trained by comparing image embeddings. We obtain competitive performance compared to ImageNet pre-training on a variety of classification datasets, from different domains. On COCO, when pre-training solely using COCO images, the detection and instance segmentation performance surpasses the supervised ImageNet pre-training in a comparable setting.

Bio

Alaa El-Nouby is a PhD student at Meta AI Research and Inria Paris advised by Hervé Jégou and Ivan Laptev. His research interests are metric learning, self-supervised learning and transformers for computer vision. Prior to pursuing his PhD, Alaa received his Msc from the University of Guelph and the Vector institute, advised by Graham Taylor, where he conducted research in spatio-temporal representation learning and text-to-image synthesis with generative models.

Tackling the Generative Learning Trilemma with Accelerated Diffusion Models
by Doctor Arash Vahdat
Date: Thursday, Feb. 10
Time: 17:30
Location: Online Call via Zoom

Our guest speaker is Arash Vahdat from NVIDIA Research and you are all cordially invited to the CVG Seminar on Feb 10th at 5:30 p.m. on Zoom (passcode is 908626).

Abstract

A wide variety of deep generative models has been developed in the past decade. Yet, these models often struggle with simultaneously addressing three key requirements including high sample quality, mode coverage, and fast sampling. We call the challenge imposed by these requirements the generative learning trilemma, as the existing models often trade between them. Particularly, denoising diffusion models (DDMs) have shown impressive sample quality and diversity, but their expensive sampling does not yet allow them to be applied in many real-world applications. 

In this talk, I will cover our two recent works on reformulating DDMs specifically for fast sampling. In the first part, I will present LSGM, a framework that allows training DDMs in a latent space. In this work, we show that by mapping data to a latent space, we can learn smoother generative processes in a smaller space, resulting in fewer network evaluations and faster sampling. In the second part, I will present denoising diffusion GANs that model the denoising distribution in DDMs using conditional GANs. In this work, we show that our multi-modal denoising distributions, in contrast to unimodal Gaussian distributions, can reduce the number of denoising steps in DDMs to as few as two steps.
 
Project Pages

https://nvlabs.github.io/LSGM/

https://nvlabs.github.io/denoising-diffusion-gan/

Bio

Arash Vahdat is a principal research scientist at NVIDIA research specializing in machine learning and computer vision. Before joining NVIDIA, he was a research scientist at D-Wave Systems where he worked on deep generative learning and weakly supervised learning. Prior to D-Wave, Arash was a research faculty member at Simon Fraser University (SFU), where he led research on deep video analysis and taught graduate-level courses on big data analysis. Arash obtained his Ph.D. and MSc from SFU under Greg Mori’s supervision working on latent variable frameworks for visual analysis. His current areas of research include deep generative learning, weakly supervised learning, efficient neural networks, and probabilistic deep learning.

Tackling the Challenge of Uncertainty Estimation and Robustness to Distributional Shift in Real-World applications
by Doctor Andrey Malinin
Date: Friday, Jan. 14
Time: 14:30
Location: Online Call via Zoom

Our guest speaker is Andrey Malinin from Yandex Research and you are all cordially invited to the CVG Seminar on Jan 14th at 2:30 p.m. on Zoom (passcode is 979103).

Abstract

While much research has been done on developing methods for improving robustness to distributional shift and uncertainty estimation, most of these methods were developed only for small-scale regression or image classification tasks. Limited work has examined developing standard datasets and benchmarks for assessing these approaches. Furthermore, many tasks of practical interest have different modalities, such as tabular data, audio, text, or sensor data, which offer significant challenges involving regression and discrete or continuous structured prediction. In this work, we propose the Shifts Dataset for evaluation of uncertainty estimates and robustness to distributional shift. The dataset, which has been collected from industrial sources and services, is composed of three tasks, with each corresponding to a particular data modality: tabular weather prediction, machine translation, and self-driving car (SDC) vehicle motion prediction. All of these data modalities and tasks are affected by real, `in-the-wild' distributional shifts and pose interesting challenges with respect to uncertainty estimation. We hope that this dataset will enable researchers to meaningfully evaluate the plethora of recently developed uncertainty quantification methods, assessment criteria and baselines, and accelerate the development of safe and reliable machine learning in real-world risk-critical applications.

An additional challenge to uncertainty estimation in real world tasks is that standard approaches, such as model ensembles, are computationally expensive. Ensemble Distribution Distillation (EnDD) is an approach that allows a single model to efficiently capture both the predictive performance and uncertainty estimates of an ensemble. Although theoretically principled, this work shows that the original Dirichlet log-likelihood criterion for EnDD exhibits poor convergence when applied to large-scale tasks where the number of classes is very high. Specifically, we show that in such conditions the original criterion focuses on the distribution of the ensemble tail-class probabilities rather than the probability of the correct and closely related classes. We propose a new training objective which resolves the gradient issues of EnDD and enables its application to tasks with many classes, as we demonstrate on the ImageNet, LibriSpeech, and WMT17 En-De datasets containing 1000, 5000, and 40,000 classes, respectively.

Bio

Andrey is a Senior Research Scientist at Yandex Research in Moscow, Russia. Prior to that He completed his PhD in Uncertainty Estimation and Speech Processing at Cambridge University under the supervision of Professor Mark Gales. His primary research interest is Bayesian-inspired approaches for Uncertainty estimation for Deep Learning and their practical application at-scale to tasks in NLP, NMT, Speech and computer vision. He also uses generative neural models, such as Flows, Variational Auto-encoders and Generative Adversarial Networks to create digital generative neural art.

(Conditional) image generation with high-degree polynomial expansions
by Grigorios Chrysos
Date: Friday, Dec. 3
Time: 14:30
Location: NS302

Our guest speaker is Grigorios Chrysos from EPFL and you are all cordially invited to the CVG Seminar on Dec 3rd at 2:30 p.m. on Zoom (passcode is 825054) or in-person (room 302 at the institute of informatics)

Abstract

Despite the impressive performance of Neural Networks (NNs), there are alternative classes of functions that can obtain similar approximation performance. In this talk, we will focus on Polynomial Networks (PNs), which use high-degree polynomial expansions to approximate the target function. The unknown parameters of PNs can be naturally represented as high-order tensors. We will exhibit how tensor decompositions can both reduce the number of learnable parameters and transform PNs into simple recursive formulations. In the second part of the talk, we will extend PNs for conditional tasks where we have multiple (possibly diverse) inputs. We will exhibit how PNs have been used for learning generative models on image, audio and non-euclidean signals. Lastly, we will showcase how conditional PNs can be used for recovering missing attribute combinations from the training set, e.g. in image generation.

Bio

Grigorios Chrysos is a Post-doctoral researcher at Ecole Polytechnique Federale de Lausanne (EPFL) following the completion of his PhD at Imperial College London (2020). Previously, he graduated from National Technical University of Athens with a Diploma/MEng in Electrical and Computer Engineering (2014). He has co-organised workshops in top conference venues, e.g. CVPR/ICCV, on deformable models. His current research interests lie in machine learning and its interface with computer vision. In particular, he is working on generative models, tensor decompositions and modelling high dimensional distributions with polynomial expansions. His recent work has been published in top tier conferences (CVPR, ICML, ICLR, NeurIPS) and prestigious journals (T-PAMI, IJCV, T-IP, Proceedings of the IEEE). He is also serving as a reviewer for the aforementioned conferences and journals. 

Understanding the Visual World through Code
by Professor Jiajun Wu
Date: Friday, Nov. 5
Time: 17:30
Location: Online Call via Zoom

Our guest speaker is Jiajun Wu from Stanford University and you are all cordially invited to the CVG Seminar on Nov 5th at 5:30 p.m. on Zoom (passcode is 769629).

Abstract

Much of our visual world is highly regular: objects are often symmetric and have repetitive parts; indoor scenes such as corridors often consist of objects organized in a repetitive layout. How can we infer and represent such regular structures from raw visual data, and later exploit them for better scene recognition, synthesis, and editing? In this talk, I will present our recent work on developing neuro-symbolic methods for scene understanding. Here, symbolic programs and neural nets play complementary roles: symbolic programs are more data-efficient to train and generalize better to new scenarios, as they robustly capture high-level structure; deep nets effectively extract complex, low-level patterns from cluttered visual data. I will demonstrate the power of such hybrid models in three different domains: 2D image editing, 3D shape modeling, and human motion understanding.

Bio

Jiajun Wu is an Assistant Professor of Computer Science at Stanford University, working on computer vision, machine learning, and computational cognitive science. Before joining Stanford, he was a Visiting Faculty Researcher at Google Research. He received his PhD in Electrical Engineering and Computer Science at Massachusetts Institute of Technology. Wu's research has been recognized through the ACM Doctoral Dissertation Award Honorable Mention, the AAAI/ACM SIGAI Doctoral Dissertation Award, the MIT George M. Sprowls PhD Thesis Award in Artificial Intelligence and Decision-Making, the 2020 Samsung AI Researcher of the Year, the IROS Best Paper Award on Cognitive Robotics, and faculty research awards and graduate fellowships from Samsung, Amazon, Facebook, Nvidia, and Adobe.