Fellowship - Topic 5: Reliable AI Contributing to Trustworthy Infrastructure Services
Japan

Odsylacz dla stanowiska: 000137

Lokalizacja: Japan

Oferta wazna do: 15/03/2021

Data publikacji oferty: 09/12/2020

Wynagrodzenie: C10M JPY

Typ zatrudnienia: Toshiba Fellowship

Rodzaj dzialalnosci: Research & Development

Topic 5: Reliable AI Contributing to Trustworthy Infrastructure Services

Background

AI technologies are remarkably advanced due to development of deep learning, so it is certain that AI will act some of the operations and maintenance of infrastructure such as electricity, transportation, and logistics, which are currently performed by humans. In Japan, the production workforce in 2030 is expected to decrease by 10% compared to 2018, and improving the efficiency of infrastructure O&M is a significant issue for a future sustainable society. Since this trend is also a big business opportunity, Toshiba has announced moves toward further growth as an infrastructure services company.

Toshiba provides various operation and/or maintenance services for infrastructures such as power generation, water and sewage, highways and railways and contributes to developing the related industries and improving quality of life. These infrastructures play a very important role in human society, so the introduction of AI cannot allow them to become unstable. Therefore, to let AI act an important and responsible role, it does not only provide high performance for specific tasks but also obtain reliability for humans by combining explainability, security and fairness. These themes are getting big attention at academic societies but further research is expected at the view of industrial applications.

Suggested Research Proposals

Research proposals could approach this from each viewpoint of explainability, security and fairness or an effort that integrates them. The points from each viewpoint are as follows.

  • Explainability: It is one of the hot topics on AI research. Though there are many studies that simply visualise the relationship between features and models, it is unclear whether they really "explain". We expect an industrially convincing explanation.
  • Security: Understanding defense against typical attacks on AI such as causing malfunction during inference by unexpected inputs, polluting the model by mixing malicious learning data.
  • Fairness: How to overcome data bias in machine learning and eliminate or suppress unfair discrimination (racism/ethnic, gender, cultural/regional discrimination, etc.).

Related Papers

[1] TOSHIBA Cyber Physical Systems; https://www.toshiba.co.jp/iot/index_en.htm
[2] Toyota Vehicles with Toshiba’s Advanced Image Recognition Processor Win Japan’s Highest Preventive Safety Performance Awards for Second Year https://toshiba.semicon-storage.com/ap-en/company/news/news-topics/2020/06/automotive-20200612-1.html
[3] Opportunities and Challenges in Explainable Artificial Intelligence (XAI): A Survey, arXiv:2006.11371, 2020.
[4] A Survey on Bias and Fairness in Machine Learning, arXiv:1908.09635, 2019.


 

Wstecz