[一般の方へ ] 学会からのお知らせ

2021年度 日本神経科学学会奨励賞受賞者 Dr. Aurelio Cortese
(Awardee of the 2021 Japan Neuroscience Society Young Investigator Award)

Metacognition, confidence and the brain's ability to learn from a small sample.

Aurelio Cortese
Computational Neuroscience Labs, ATR Institute International
I am very honored to receive the Japan Neuroscience Society Young Investigator Award, particularly as the first foreigner in the history of the award. More importantly, I would like to take this opportunity to express my gratitude to the selection committee, to my mentors and many collaborators who have guided me and made my scientific journey so enjoyable.
As a child, I loved experimenting new things, and I was fascinated by the nature of our sensory experiences. The sensation of the sun on the skin in a warm spring afternoon, or the smell of flowers in a grass field, or the sharp pain of a wasp bite. As I grew up and studied biology, I wondered how can billions of neurons and electrical signals give rise to our awareness, thoughts and cognition? Thus, I began my journey to understand the inner workings of our mind.
From my PhD studies at NAIST, under the supervision of Profs. Kazushi Ikeda, Kaoru Amano (Cinet), Hakwan Lau (UCLA) and Mitsuo Kawato (ATR), I studied metacognition and decision-making in human subjects, using neuroimaging (fMRI) and machine learning. At the time, I sought to clarify an old debate in the literature: whether our feeling of confidence, when we make choices, is a simple measure of the information available, or whether it is a higher order ‘meta-cognitive’ process. I used a new technique, called ‘decoded neurofeedback’, to change a person’s confidence at the neural level, non-invasively with fMRI (Nature Commun 2016). Looking further ahead, the decoded neurofeedback approach seemed incredibly promising: if we can change mental representations non-invasively with fMRI, there may be powerful clinical innovations. For example, to treat phobia or PTSD by remapping fearful memories (Nature Hum Behav 2016, PNAS 2018).
But as I progressed, I also realized that decoded neurofeedback presented an incredible theoretical challenge. How does the brain learn to shape its own activity patterns in response to external teaching signals? This is particularly mysterious because in these experiments the feedback is based on fMRI rather than direct neural signals, and the brain does not know which area nor what ‘information’ are targeted. During my time as postdoctoral researcher I embarked on a series of studies to investigate the function of cognition in learning very complex problems from a small sample (i.e., without extensive experience). I first discovered that decision confidence accelerates reinforcement learning, and that it does so via the coupling between the dorsal part of the prefrontal cortex and the basal ganglia (Nature Commun. 2020).
Increasingly fascinated by the flexibility of biological learning and decision-making, I have continued to expand my research in computational neuroscience towards other cognitive dimensions (abstraction, attention, memory), and psychiatric disorders (PTSD, depression, anxiety). Since 2019, I have been able to develop a multidisciplinary team to clarify the nature of flexible and adaptive behaviors. From this point of view, I am particularly interested in exploring the intersection between neuroscience and artificial intelligence (AI). Computational models in neuroscience can be used to implement new algorithms in AI that are more efficient and ethical. At the same time, techniques or methods from these fields can provide new ways to analyze and understand neural data. We may also open new ways in which humans and AI can interact, with a positive impact on society.
Looking back, I would like to take this opportunity to express my deep gratitude to Professor Mitsuo Kawato for helping me grow as a researcher over the years, and giving me the freedom to explore many ideas. I wouldn't be here without the guidance of Professor Hakwan Lau and Professor Benedetto De Martino, and the encouragement of the many other senior scientists I have encountered and discussed with. Special thanks to Professor Yuji Ikegaya for his trust and giving me the opportunity to contribute innovative research at the intersection of neuroscience and AI in the ERATO Ikegaya Brain-AI Hybrid project. Finally, special mention to all the students I worked with and who devoted their enthusiasm to this research.
* Review article on the content of the awarded research (Published in Neuroscience Research)
Cortese, A., 2021. Metacognitive resources for adaptive learning. Neurosci. Res. 178, 10-19
* Brief sketch of my career
2007 – 2010 BSc in Life Science & Technology, EPFL, Switzerland
2010 – 2012 MSc in Life Science & Technology, EPFL, Switzerland
2011 – 2012 Visiting researcher, Max-Planck Institute for Psychiatry, Germany
2013 – 2016 PhD Information Science / Neuroscience, NAIST, Japan
2015 07 – 12 Visiting graduate student, UCLA, US
2017 – 2018 Postdoctoral researcher, ATR Computational Neuroscience Labs, Japan
2018 – pres. Principal Investigator, ERATO Ikegaya Brain-AI Hybrid project
2019 – 2021 Senior Researcher, ATR Computational Neuroscience Labs, Japan
2019 – pres. Visiting Researcher, BDM Lab, University College London, UK
2021 – pres. Chief Researcher / Decoded Neurofeedback Vice-Department Head, ATR Computational Neuroscience Labs, Japan
Aurelio Cortese (Computational Neuroscience Labs, ATR Institute International)



ATR 脳情報研究所
奈良先端科学技術大学院大学の博士課程では、池田和司教授、天野薫氏(NICT)、Hakwan Lau教授(UCLA)、川人光男所長(ATR)の指導のもと、神経画像(fMRI)と機械学習を用いて、メタ認知と意思決定に関する研究を行いました。当時、我々が下す選択に関する確信度という感覚が、単に選択の判断材料となる情報量に依存するのか、高次の「メタ認知」を含むプロセスなのか、という論争が古くから続いていました。この論争を解決するため、私は「デコーディドニューロフィードバック」と呼ばれるfMRIを用いた非侵襲的な最先端の介入手法を用いて、神経レベルで人の確信度を変化させました(Nature Commun 2016)。デコーディドニューロフィードバックによるアプローチは、将来性の高い、非常に有望な手法であると感じました。精神状態に関する神経表象をfMRIで非侵襲的に変化させることができれば、臨床分野に革命をもたらすことが期待されます。例えば、恐怖の記憶に関する神経表象を書き換えることで、恐怖症やPTSDを治療することも可能になります(Nature Hum Behav 2016, PNAS 2018)。
しかし、研究を進めていくにつれて、ニューロフィードバックが理論的に非常に難しい課題を内包していることに気づきました。脳はどのようにして外部の教示信号に反応して自らの活動パターンを形成することができるのでしょう。問題を複雑にしているのが、この手法が実際の神経信号ではなくfMRI信号に基づいたフィードバックを行うという点や、脳がどこのどのような「情報」がターゲットになっているのか分からないとう点です。これらの問題に取り組むため、ポスドク時代には、少数のサンプル(つまり、豊富な経験がない状態)から非常に複雑な問題を学習する際の認知機能に関する一連の研究に着手しました。これにより、意思決定の確信度が強化学習を加速させること、そしてそれが前頭前野の背側部分と大脳基底核の間の結合の強化を介して行われることを発見しました (Nature Commun. 2020)。基底核の間の結合を介して行われることを発見しました (Nature Commun. 2020)。
最後になりましたが、長年にわたって私を研究者として成長させ、さまざまなアイデアを自由に探求させてくれた川人光男教授にこの場を借りて深く感謝いたします。また、Hakwan Lau教授、Benedetto De Martino教授のご指導や、これまでに出会った多くの研究者の方々の励ましがなければ、今の私はありません。ERATO池谷脳AI融合プロジェクトにおいて、神経科学とAIの融合という革新的な研究に貢献する機会を与えて下さった池谷裕二教授に深く感謝いたします。そして共に研究を行い、研究へ多くの情熱を注いでくれたすべての学生へも感謝の意を表します。