Home

Welcome

Welcome to the webpage of the Adaptive Information Processing Group. The group focuses on information processing schemes that adapts to the environment and situation, changing the algorithms and/or the models adaptively. Humans and all living beings have wonderful abilities to solve various problems by trial and error, sometimes spending generations, by using a very limited information processing resource. The main research field of this group is to solve real-world problems, by methods based on learning, optimization and signal processing, devising new algorithms for pattern recognition, signal/image processing and retrieval; the scheme we call the Adaptive Information Processing.

Call for graduation research members 2025 & Open Lab info

(This announcement is intended for undergraduate students enrolled in Univ. Tsukuba. )

Come to study and implement adaptive information processing abilities found in humans and nature!. All students with fresh ideas and enthusiasm are welcomed.

Open laboratory (2024) :

Wednesday Oct.2th 17:00-18:00 @3E102-1
Thursday Oct. 17th 17:00-18:00 @3E102-1
Friday Oct. 25th 17:00-18:00 @3E102-1

Please send to info *at* adapt.cs.tsukuba.ac.jp if you have any questions.
Visits welcomed at any time. Please contact us.

Papers presented at IEEE CSPA 2023

The following works have been presented at the 20th IEEE International Colloquium on Signal Processing & Its Applications, held during 3rd-4th March 2023 at Langkawi , Malaysia.

Pavodi Ndoyi Maniamfu and Keisuke Kameyama
“LSTM-based forecasting using policy stringency and time-varying parameters of the SIR model for COVID-19”
Tzu-Jui Huang and Keisuke Kameyama
“Machine Learning Curriculums Generated by Classifier Ensembles”
Takumi Morikawa and Keisuke Kameyama
“CNN Model Compression by Merit-Based Distillation” (Awarded best paper at CSPA)

Feature selection based on joint-conditional mutual information

A novel method for feature channel selection of the hyperspectral images obtained in remote-sensing, was presented in IEEE Symposium Series on Computational Intelligence, held at Singapore Management University during Dec. 4-7, 2022.

U. A. Md. Ehsan Ali and Keisuke Kameyama, “Informative Band Subset Selection for Hyperspectral Image Classification using Joint and Conditional Mutual Information,” IEEE Symposium on Computational Intelligence in Remote Sensing (IEEE SSCI 2022), (Singapore), pp. 573-580, Dec. 2022.

The paper proposes a novel feature selection method (JCIF) based on joint-conditional mutual information between the hyperspectral channels.The method was applied to toy pattern recognition problems and the segmentation of real hyperspectral remote sensing data, and it was shown to select the feature channels which enable a superior classification ratio in comparison with the other known feature selection methods.

Paper on multi-stage neural network model compression presented at PerConAI

The paper on convolutional neural network model compression, jointly using distillation and hint-based learning in multiple stages,

Takumi Morikawa and Keisuke Kameyama

"Multi-Stage Model Compression using Teacher Assistant and Distillation with Hint-Based Training",

was presented in the First Workshop on Pervasive and Resource-Constrained Artificial Intelligence (PerConAI). The workshop was held in part of The 20th IEEE International Conference on Pervasive Computing and Communications (PerCom 2022), on March 25, 2022.

Paper on multimodal person recognition presented at ICONIP 2021

Our paper

Keita Ogawa and Keisuke Kameyama

"Adaptive Selection of Classifiers for Person Recognition by Iris Pattern and Periocular Image",

which proposes the Multi Modal Selector for adaptively selecting a reliable modal classifier in multimodal person identification, was presented in the oral session of the International Conference on Neural Information Processcing (ICONIP) 2021 held during Dec. 8-11, 2021.