Wyniki wyszukiwania - Biblioteka Nauki

1

Handling realistic noise in multi-agent systems with self-supervised learning and curiosity

100%

Szemenyei M. , Reizinger P.

Journal of Artificial Intelligence and Soft Computing Research

|

2022

|

tom Vol. 12, No. 2

135--148

EN

Most reinforcement learning benchmarks – especially in multi-agent tasks – do not go beyond observations with simple noise; nonetheless, real scenarios induce more elaborate vision pipeline failures: false sightings, misclassifications or occlusion. In this work, we propose a lightweight, 2D environment for robot soccer and autonomous driving that can emulate the above discrepancies. Besides establishing a benchmark for accessible multiagent reinforcement learning research, our work addresses the challenges the simulator imposes. For handling realistic noise, we use self-supervised learning to enhance scene reconstruction and extend curiosity-driven learning to model longer horizons. Our extensive experiments show that the proposed methods achieve state-of-the-art performance, compared against actor-critic methods, ICM, and PPO.

2

Self-supervised learning of motion-induced acoustic noise awareness in social robots

86%

Andrade J. , Pedro S. , Almeida A. P.

Journal of Automation Mobile Robotics and Intelligent Systems

|

2019

|

tom Vol. 13, No. 1

3--14

EN

With the growing presence of robots in human populated environments, it becomes necessary to render their presence natural, rather than invasive. To do that, robots need to make sure the acoustic noise induced by their motion does not disturb people nearby. In this line, this paper proposes a method that allows the robot to learn how to control the amount of noise it produces, taking into account the environmental context and the robot’s mechanical characteristics. Concretely, the robot adapts its motion to a speed that allows it to produce less noise than the environment’s background noise and, hence, avoiding to disturb nearby humans. For that, before executing any given task in the environment, the robot learns how much acoustic noise it produces at different speeds in that environment by gathering acoustic informatinon through a microphone. The proposed method was successfully validated on various environments with various background noises. In addition, a PIR sensor was installed on the robot in order to test the robot’s ability to trigger the noise-aware speed control procedure when a person enters the sensor’s field of view. The use of a such a simple sensor aims at demonstrating the ability of the proposed system to be deployed in minimalistic robots, such as micro unmanned aerial vehicles.

3

A seismic random noise suppression method based on self-supervised deep learning and transfer learning

72%

Wu T. , Meng X. , Liu H. , Li W.

Acta Geophysica

|

2024

|

tom Vol. 72, no. 2

655--671

EN

Random noise suppression is an essential task in the seismic data processing. In recent years deep learning methods have achieved superior results in seismic data denoising. However, obtaining clean data from field seismic data for training is challenging. Therefore, supervised deep learning denoising methods can only use synthetic datasets or field datasets constructed by conventional seismic denoising methods for training. Aiming at this problem, we proposed a self-supervised deep learning seismic denoising method based on Neighbor2Neighbor. This method only requires sampling the noisy data twice to train the denoising network without clean data. For the characteristics of seismic data, we designed a vertical neighbor subsample to make Neighbor2Neighbor more suitable for seismic data. In addition, to further improve the denoising effect in field data, we introduced a transfer learning strategy in our method. Numerical experiments demonstrated that our method outperformed both the conventional denoising seismic method and the supervised learning seismic denoising method after transfer learning.

4

Can Unlabelled Data Improve AI Applications? A Comparative Study on Self-Supervised Learning in Computer Vision

58%

Bauer M. , Augenstein C.

Annals of Computer Science and Information Systems

|

2023

|

tom Vol. 35

93--101

EN

Artificial Intelligence (AI) represents a highly investigated area of study at present and has already become an indispensable component within an extensive range of business models and applications. One major downside of current supervised AI approaches lies in the need of numerous annotated data points to train the models. Self-supervised learning (SSL) circumvents the need for annotation, by creating supervision signals such as labels from the data itself, rather than requiring experts for this task. Current approaches mainly include the use of generative methods such as autoencoders and joint embedding architectures to fulfil this task. Recent works present comparable results to supervised learning in downstream scenarios such as classification after SSL-pretraining. To achieve this, typically modifications are required to suit the approach for the exact downstream task. Yet, current review works haven't paid too much attention to the practical implications of using SSL. Thus, we investigated and implemented popular SSL approaches, suitable for downstream tasks such as classification, from an initial collection of more than 400 papers. We evaluate a selection of these approaches under real-world dataset conditions, and in direct comparison to the supervised learning scenario. We conclude that SSL has the potential to take up with supervised learning, if the right training methods are identified and applied. Furthermore, we also introduce future directions for SSL research, as well as current limitations in real-world applications.