Using Opaque AI for Smart Grids
(2020) SYSK16 20201Department of Informatics
- Abstract
- Deep learning is an emerging machine learning technique which can find complex patterns in large amounts of data. This makes it useful for several applications in smart grids, which often involve the processing of large amounts of data. However, there are reasons to be sceptical of its suitability as the black-box nature of deep learning could be a problem since power grids are important infrastructure and contain deadly currents. Professionals in smart grids were interviewed to provide an understanding of the importance of eight issues relating to the interpretability of machine learning. The findings show that for some uses related to controlling the grid, trust is of critical importance, and it is unlikely that a black-box algorithm... (More)
- Deep learning is an emerging machine learning technique which can find complex patterns in large amounts of data. This makes it useful for several applications in smart grids, which often involve the processing of large amounts of data. However, there are reasons to be sceptical of its suitability as the black-box nature of deep learning could be a problem since power grids are important infrastructure and contain deadly currents. Professionals in smart grids were interviewed to provide an understanding of the importance of eight issues relating to the interpretability of machine learning. The findings show that for some uses related to controlling the grid, trust is of critical importance, and it is unlikely that a black-box algorithm will be used. For other uses such as giving recommendations and forecasts, it was found that either trust or informativeness is required for the results to be useful, although trust could potentially be achieved through a strong track-record, rather than through the ability to interpret the system. Other issues were of varying importance, but none of them critical. Unless the area of interpretability sees considerable progress, it will be of concern when creating deep learning systems for smart grids. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9016702
- author
- Lundberg, Jesper LU and Lundborg, Alexander LU
- supervisor
- organization
- alternative title
- How does the difficulty of interpreting deep learning systems affect their suitability for smart grids?
- course
- SYSK16 20201
- year
- 2020
- type
- M2 - Bachelor Degree
- subject
- keywords
- Deep Learning, Smart Grids, Interpretability, Black-Box, Transparency, Post- hoc Explainability, AI, Machine Learning
- report number
- INF20-01
- language
- English
- id
- 9016702
- date added to LUP
- 2020-06-26 14:06:14
- date last changed
- 2020-06-26 14:06:14
@misc{9016702, abstract = {{Deep learning is an emerging machine learning technique which can find complex patterns in large amounts of data. This makes it useful for several applications in smart grids, which often involve the processing of large amounts of data. However, there are reasons to be sceptical of its suitability as the black-box nature of deep learning could be a problem since power grids are important infrastructure and contain deadly currents. Professionals in smart grids were interviewed to provide an understanding of the importance of eight issues relating to the interpretability of machine learning. The findings show that for some uses related to controlling the grid, trust is of critical importance, and it is unlikely that a black-box algorithm will be used. For other uses such as giving recommendations and forecasts, it was found that either trust or informativeness is required for the results to be useful, although trust could potentially be achieved through a strong track-record, rather than through the ability to interpret the system. Other issues were of varying importance, but none of them critical. Unless the area of interpretability sees considerable progress, it will be of concern when creating deep learning systems for smart grids.}}, author = {{Lundberg, Jesper and Lundborg, Alexander}}, language = {{eng}}, note = {{Student Paper}}, title = {{Using Opaque AI for Smart Grids}}, year = {{2020}}, }