Dmytro Kuzmenko

May 9

CLIP-Gen Overview

Intro Training a text-to-image generator in the general domain like DALL-E, GauGAN, and CogView requires huge amounts of paired text-image data, which can be problematic and expensive. In this paper, the authors propose a self-supervised scheme named CLIP-GEN for general text-to-image generation with the language-image…

Deep Learning

6 min read

Apr 29

Overview — ZeroGen, Efficient Zero-shot Learning via Dataset Generation

An interesting take on zero-shot learning was introduced in a paper that was dated Feb 16. Some preliminary knowledge about prompt-based zero-shot learning framework: Given a pre-trained language model (PLM), and a text classification task, Prompting first instantiates a prompt with each input and outputs a natural language sequence to…

Zero Shot Learning

2 min read

Apr 21

Overview — AutoDIME, Automatic Design of Interesting Multi-Agent Environments

Background Finding the precise environmental conditions that lead to interesting emergent skills in complex environments is often a time-consuming and laborious process. One strategy to mitigate this problem is to train agents on a wide distribution of randomised environments, in the hope that some small fraction of them will lead to…

Reinforcement Learning

5 min read

Overview — AutoDIME, Automatic Design of Interesting Multi-Agent Environments

Mar 24

Overview — [2022–03–14] LiDAR-based 4D Panoptic Segmentation via Dynamic Shifting Network

TL;DR In this work, the authors address the task of LiDAR-based panoptic segmentation, which aims to parse both objects and scenes in a unified manner. Dynamic Shifting Network (DS-Net) is proposed, which serves as an effective panoptic segmentation framework in the point cloud realm. DS-Net has 3 appealing properties: Strong backbone…

Autonomous Driving

8 min read

Overview — [2022–03–14] LiDAR-based 4D Panoptic Segmentation via Dynamic Shifting Network

Mar 21

Overview — [2022–03–11] Block-Recurrent Transformer

Block-Recurrent Transformers The authors introduce the Block-Recurrent Transformer, which applies a transformer layer in a recurrent fashion along a sequence, and has linear complexity with respect to the sequence length. The recurrent cell operates on blocks of tokens rather than single tokens and leverages parallel computation within a block in order to…

Transformers

10 min read

Mar 16

Overview — [2021–06–24] Video Swin Transformer

Intro Video Swin Transformer advocates an inductive bias of locality in video Transformers, leading to a better speed-accuracy trade-off compared to previous approaches which compute self-attention globally even with spatial-temporal factorization. The locality of the proposed video architecture is implemented by adapting the Swin Transformer designed for the image domain while…

Deep Learning

9 min read

Mar 13

Overview — [2021–12–09] LipSound2: Self-Supervised Pre-Training for Lip-to-Speech Reconstruction and Lip Reading

Foreword Greetings! I am further expanding my AI/ML-related news sharing, detailed paper reviews, guest lectures, and future scientific works with engineers and researchers who are already entangled in the world of AI, specialists from other domains looking to delve deeper into Machine Learning, and everyone keen on the topic and wanting…

Deep Learning

5 min read

Overview — LipSound2: Self-Supervised Pre-Training for Lip-to-Speech Reconstruction and Lip Reading