Artificial intelligence (AI), in particular Deep learning, has since the ImageNet LSVRC-2012 contest established itself as one of the core technologies driving the 4th industrial revolution with many commercial applications. This rapid success of Artificial narrow intelligence (ANI) is due to four factors: big labelled data, GPU accelerated distributed computing, open source software, and algorithms. Labelled training data for many diseases, such as Alzheimer’s disease (AD), is however scarce and costly. Humans can learn from a single example, while artificial neural networks (ANNs) require orders of magnitude more samples. Currently, Taiwan is lagging behind in the adoption and development of AI.

We intend to put Taiwan on the global AI map through application driven basic research focused on creation of new learning methods for ANNs inspired by how children learn. We will track the complexity of the sensorimotor function and learning process in children based on sensors embedded in toys during play at home. We plan to obtain data almost daily, following each child from roughly 3- months until 3-years of age. We plan to use System identification techniques to derive dynamical models within the dynamical systems theory framework and Deep learning to train neural networks within the connectionist framework as models for explaining cognitive development. We also plan to test established system and cellular/microcircuit level models of neuronal mechanisms developed by our international collaborator Dr. Hellgren Kotaleski at KTH Royal Institute of Technology/Karolinska Institutet in Sweden. These models will likely be the first dynamical models of sensorimotor function development in individual children.

Based on these insights, we intend to invent a learning strategy that decreases the number of samples needed for learning of deep convolutional neural networks (DCNNs), so that we can predict onset of AD from neuroimaging data before signs of mild cognitive impairment. According to the amyloid hypothesis, increased levels of the amyloid-β protein in the brain lead to damage and dysfunction of affected synapses and neurons causing dementia. Since this damage so far has proven irreparable, early detection of onset is thought to be the key to treatment. We intend to use magnetic resonance imaging (MRI) data from ADNI and our collaborators in Taiwan and U.S.A. to train a DCNN to predict onset of AD and changes in cognitive ability. Through this cross-disciplinary research we will attract and train >20 domestic and international students as AI professionals. Together we plan to form a startup to commercialise the innovations for the benefit of the society.

Keyword： Artificial intelligence (AI), Artificial neural networks (ANNs), Deep learning, Artificial narrow intelligence (ANI), Alzheimer’s disease (AD), Systems biology, Bioinformatics, Reverse engineering, System identification

AI於生醫之認知雲端與神經型態運算–深度學習與人及機器的互動對阿滋海默的早期發現

自從ImageNet LSVRC-2012競賽以來，人工智慧，尤其是深度學習成為工業4.0的重要元素。狹義人工智慧(弱人工智能)能快速的成功有四大因素：帶標籤的大數據、GPU高性能運算、開源軟體以及演算法。在疾病研究上，如阿茲海默症，其資料非常難取得，使得此研究難以推進，目前的人工神經網路仍需要大量的數據。目前台灣在AI發展這塊領域仍是遲滯不前 。 我們欲建立以應用為出發點來發展基礎研究，著重在建立一個藉由人類幼兒的學習方式來啟發人工神經網路的新學習方法。期待此計畫能將台灣推向全球網絡。我們將追蹤3個月至3歲大的孩童獲取近每天之數據。我們計畫使用系統識別技術推導出動態系統理論體中的動態模組與深度學習，以聯結主義為模型訓練神經網路，來解釋認知發展。我們也將測試我們國際合作團隊：瑞典皇家工學院的Hellgren Kotaleski博士發展的既設系統以及神經機制之細胞/微電路尺度模型。這個得到的模組，很有潛力成為第一個應用在幼兒感覺運動功能發展上的動態模組。 基於上述概念，我們試圖創造一個能夠減少卷積神經網路所需樣本數的學習戰略，以利於我們可以從神經影像 (neuroimaging) 預測在有輕微辨識障礙前的阿茲海默症病友的發病時間。根據類澱粉蛋白假說，當腦中的β-類澱粉蛋白增加會導致突觸及神經元損傷以致引發癡呆症，此損傷至今證實是不可修復的，其治療的關鍵在於發病前的早期預測。我們將使用ADNI資料庫以及我們與台灣、美國、加拿大以及瑞典實驗團隊合作的核磁共振造影資料，來訓練深度卷積神經網路以預測阿茲海默症的發病時間以及認知能力的改變。經由這個跨領域研究，未來能吸引到國內外超過二十名國內外學生並將他們培育成人工智慧方面的專家，此外，我們預計將創新創業，將這個創新理念商品化，為社會帶來貢獻。

關鍵字： 人工智能（AI）、人工神經網絡（ANN）、深度學習、弱人工智能（ANI）、阿茲海默病（AD）、系統生物學、生物資訊學 、反向工 程 、系統辨識

Principal Investigator: Ass. Prof. Torbjörn Nordling. 

Co-principal Investigator: Prof. Yu-Min Kuo, Dept. of Cell Biol. and Anatomy, National Cheng Kung University.

Sponsor: Ministry of Science and Technology in Taiwan 

Grant identifiers: 107-2634-F-006-009, 108-2634-F-006-009.

Duration: 2018-01-01 to 2019-07-31.

Collaborators: 1. Erik Sonnhammer, Prof., Dept. of Biochemistry and Biophysics, Stockholm University, Sweden. 

2. Jeanette Hellgren Kotaleski, Prof., Dept. of Computational Sci. and Tech., School of Comp. Sci. and Com., KTH 
Royal Institute of Technology, and Dept. of Neuroscience, Karolinska Institutet, Stockholm, Sweden.
3. Filippo Menolascina, Chancellor’s Fellow, School of Engineering, University of Edinburgh, U.K. 

4. Chris Lucas, Chancellor’s Fellow, School of Informatics, Univ. of Edinburgh, U.K. 

5. Patricia Brosseau-Liard, Ass. Prof., School of Psychology, University of Ottawa, Canada. 

6. Stephanie Denison, Asc. Prof., Dept. of Psych., University of Waterloo, Canada.
7. Daphna Buchsbaum, Ass. Prof., Dept. of Psych., University of Toronto, Canada.
8. David Sarne, Asc. Prof., Dept. of Computer Science, Bar-Ilan University, Israel.
9. Tammy Riklin Raviv, Asc. Prof., Dept. of Elect. Eng., Zlotowski Center for Neuroscience, Ben-Gurion University of the Negev, Israel.
10. Sarit Alkalay, Lect., Dept. of Psych., Dept. of Beh. Sci., Jezreel Valley College, and Int. M.A. program in Child Devel., Faculty of Social Sci., Univ. of Haifa, Israel. 

11. Avivit Dolev, Researcher, VR & NeuroCognition Lab, Edu. in Sci. and Tech., Israel Inst. of Tech. (Technion), Haifa, Israel. 

12. Daisuke Yamamoto, Prof. Em., PI, Adv. ICT Res. Inst., National Inst. of Info. and Communications Tech., Kobe, Japan. 

13. Margaretha, Lect., Clinical Psych. and Mental Health Dept. Universitas Airlangga, Surabaya, Indonesia.
14. Wei-Sheng “Wesson” Wu, Prof., Dept. Electrical Engineering, National Cheng Kung University, Taiwan.
15. Darsen Lu, Ass. Prof., Inst. of Microelectronics, National Cheng Kung University, Taiwan.
16. Gwo-Giun “Chris” Lee, Prof., Dept. of Elect. Eng., National Cheng Kung University, Taiwan.
17. Shulan Hsieh, Dist. Prof., Dept. of Psychology, National Cheng Kung University, Taiwan.
18. Yun-Hsuan Chang, Lic. clinical psychologist, Dept. of Psychology, Asia University, Taiwan.
19. Chun-Hao Wang, Ass. Prof., Inst. of Physical Education, National Cheng Kung University, Taiwan. 20. Che-Wei “Johnny” Lin, Ass. Prof., Dept. of BioMedical Eng., National Cheng Kung University, Taiwan.
21. Tzu-Cheng Chao, Asc. Prof., Dept. of Comp. Science and Info. Eng., National Cheng Kung University, Taiwan.
22. Chia-Fen Hsu, Ass. Prof., Department of Psychology, Chung Shan Medical University, Taiwan.
23. Mario Hofmann, Asc. Prof., Dept. of Physics, National Taiwan University, Taiwan.

Members: David Nokto, Anette Kniberg, Paul Tsai, Shun-Jie Zhang, Alan Yu, Lewis Hsu, Rain Wu, Nick Chen, Oliver Lai, Justin Lin, Butters Wu, Jack Wang, Tria Puspa Sari, Dickson Lee, Ray Chen, Jose Ramon Chang, Alexander Heimann, Esteban Roman Catafau, Kyle Liu, Heng-Ying Chou, Gina Wu, Fu-Yu Zhang, Bart van Beuren, Mark Liou, Paola di Maio, Akram Ashyani, Torbjörn Nordling.