Open in app
Sign inGet started
Latest
A Systematic Methodology for Continuous-Time Analysis of Accelerated Gradient Methods

A Systematic Methodology for Continuous-Time Analysis of Accelerated Gradient Methods

In machine learning, we often minimize the loss function so that the difference between the solution point and the estimates converges to…
Go to the profile of SNU AI
SNU AI
Check Factual Consistency of Auto-generated Summaries Using Masked Summarization

Check Factual Consistency of Auto-generated Summaries Using Masked Summarization

Key information in source texts and reference summaries is masked to artificially generate factually inconsistent summaries
Go to the profile of SNU AI
SNU AI
The Intersection of Human Bias and Artificial Intelligence

The Intersection of Human Bias and Artificial Intelligence

Understanding how and why AI reflects human bias like a mirror helps to shape balanced AI regulatory governance
Go to the profile of SNU AI
SNU AI
Measuring and Explaining the Inter-Cluster Reliability of Multidimensional Projections

Measuring and Explaining the Inter-Cluster Reliability of Multidimensional Projections

Distortions occur when reducing dimensionality. Our metrics quantitatively measure how well the low-dimensional projection preserves the…
Go to the profile of SNU AI
SNU AI
Visual Graph Memory with Unsupervised Representation for Visual Navigation

Visual Graph Memory with Unsupervised Representation for Visual Navigation

How can an agent navigate through a particular environment to reach target observation, only using visual sensory signals?
Go to the profile of SNU AI
SNU AI
Puzzle Mix: Exploiting Saliency and Local Statistics for Optimal Mixup

Puzzle Mix: Exploiting Saliency and Local Statistics for Optimal Mixup

Puzzle Mix is a data augmentation method useful for optimally leveraging the saliency information respecting the local statistics of the..
Go to the profile of SNU AI
SNU AI
Imperative-Symbolic Co-Execution of Imperative Deep Learning Programs

Imperative-Symbolic Co-Execution of Imperative Deep Learning Programs

Our approach can handle any imperative DL program while achieving the optimized performance of symbolic execution.
Go to the profile of SNU AI
SNU AI
Model-based Domain Randomization of Dynamics System with Deep Bayesian Locally Linear Embedding

Model-based Domain Randomization of Dynamics System with Deep Bayesian Locally Linear Embedding

Domain randomization is a simple technique to enhance the robustness of the policy in various environments. We aim for efficient policy…
Go to the profile of SNU AI
SNU AI
Refurbish Your Training Data for Faster DNN Training

Refurbish Your Training Data for Faster DNN Training

Data refurbishing is a novel sample reuse mechanism that accelerates DNN training while preserving model generalization.
Go to the profile of SNU AI
SNU AI
Expressive 3D Human Pose and Shape Estimation, Part 2: Mesh Estimation and 3D Rotational Pose…

Expressive 3D Human Pose and Shape Estimation, Part 2: Mesh Estimation and 3D Rotational Pose…

Here’s how we estimate the human mesh and capture all human parts, including hands and the face, for better human behavior understanding.
Go to the profile of SNU AI
SNU AI
Expressive 3D Human Pose and Shape Estimation, Part 1: Multi-person and Interacting Hand Pose

Expressive 3D Human Pose and Shape Estimation, Part 1: Multi-person and Interacting Hand Pose

Measuring a subject’s relative distance from the camera for multi-person scenarios and depicting the complex sequence of interacting…
Go to the profile of SNU AI
SNU AI
DeepCuts: a Deep Learning Optimization Framework for Versatile GPU Workloads

DeepCuts: a Deep Learning Optimization Framework for Versatile GPU Workloads

GPUs are the de facto standard to run DL applications. DeepCuts considers both kernel implementation and GPU architecture parameters to…
Go to the profile of SNU AI
SNU AI
Continual Learning with Node-wise Importance Regularization

Continual Learning with Node-wise Importance Regularization

Continual learning refers to the ability of a model to learn continually from a stream of data. If the model focuses too much on…
Go to the profile of SNU AI
SNU AI
Robustifying Multi-hop QA through Pseudo-Evidentiality Training

Robustifying Multi-hop QA through Pseudo-Evidentiality Training

Complex questions require connecting evidence from several texts. Our QA model recognizes whether its answer is supported by evidence.
Go to the profile of SNU AI
SNU AI
Learning Discrete Compressed Representation for Noise-Robust Exploration

Learning Discrete Compressed Representation for Noise-Robust Exploration

What can we do about unsuitable information in data that manipulates DNNs? Our method jointly learns features and drops task-irrelevant…
Go to the profile of SNU AI
SNU AI
Lightweight and Parallel GPU Task Scheduling for Deep Learning

Lightweight and Parallel GPU Task Scheduling for Deep Learning

It is natural to expect DL frameworks to utilize the computation power of GPU to the fullest. We design an execution engine atop PyTorch…
Go to the profile of SNU AI
SNU AI
About SNU AIIS BlogLatest StoriesArchiveAbout MediumTermsPrivacyTeams