Training self-learning circuits for power-efficient solutions
As the size and ubiquity of artificial intelligence and computational machine learning models grow, the energy required to train and use them is rapidly becoming economically and environmentally unsustainable. Recent laboratory prototypes of self-learning electronic circuits, such as “physical learn...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
AIP Publishing LLC
2024-03-01
|
| Series: | APL Machine Learning |
| Online Access: | http://dx.doi.org/10.1063/5.0181382 |
| _version_ | 1797228277696299008 |
|---|---|
| author | Menachem Stern Sam Dillavou Dinesh Jayaraman Douglas J. Durian Andrea J. Liu |
| author_facet | Menachem Stern Sam Dillavou Dinesh Jayaraman Douglas J. Durian Andrea J. Liu |
| author_sort | Menachem Stern |
| collection | DOAJ |
| description | As the size and ubiquity of artificial intelligence and computational machine learning models grow, the energy required to train and use them is rapidly becoming economically and environmentally unsustainable. Recent laboratory prototypes of self-learning electronic circuits, such as “physical learning machines,” open the door to analog hardware that directly employs physics to learn desired functions from examples at a low energy cost. In this work, we show that this hardware platform allows for an even further reduction in energy consumption by using good initial conditions and a new learning algorithm. Using analytical calculations, simulations, and experiments, we show that a trade-off emerges when learning dynamics attempt to minimize both the error and the power consumption of the solution—greater power reductions can be achieved at the cost of decreasing solution accuracy. Finally, we demonstrate a practical procedure to weigh the relative importance of error and power minimization, improving the power efficiency given a specific tolerance to error. |
| first_indexed | 2024-04-24T14:54:09Z |
| format | Article |
| id | doaj.art-066f6c10190a4390bef81dfc1b17dd47 |
| institution | Directory Open Access Journal |
| issn | 2770-9019 |
| language | English |
| last_indexed | 2024-04-24T14:54:09Z |
| publishDate | 2024-03-01 |
| publisher | AIP Publishing LLC |
| record_format | Article |
| series | APL Machine Learning |
| spelling | doaj.art-066f6c10190a4390bef81dfc1b17dd472024-04-02T19:46:06ZengAIP Publishing LLCAPL Machine Learning2770-90192024-03-0121016114016114-1610.1063/5.0181382Training self-learning circuits for power-efficient solutionsMenachem Stern0Sam Dillavou1Dinesh Jayaraman2Douglas J. Durian3Andrea J. Liu4Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USADepartment of Physics and Astronomy, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USADepartment of Computer and Information Science, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USADepartment of Physics and Astronomy, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USADepartment of Physics and Astronomy, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USAAs the size and ubiquity of artificial intelligence and computational machine learning models grow, the energy required to train and use them is rapidly becoming economically and environmentally unsustainable. Recent laboratory prototypes of self-learning electronic circuits, such as “physical learning machines,” open the door to analog hardware that directly employs physics to learn desired functions from examples at a low energy cost. In this work, we show that this hardware platform allows for an even further reduction in energy consumption by using good initial conditions and a new learning algorithm. Using analytical calculations, simulations, and experiments, we show that a trade-off emerges when learning dynamics attempt to minimize both the error and the power consumption of the solution—greater power reductions can be achieved at the cost of decreasing solution accuracy. Finally, we demonstrate a practical procedure to weigh the relative importance of error and power minimization, improving the power efficiency given a specific tolerance to error.http://dx.doi.org/10.1063/5.0181382 |
| spellingShingle | Menachem Stern Sam Dillavou Dinesh Jayaraman Douglas J. Durian Andrea J. Liu Training self-learning circuits for power-efficient solutions APL Machine Learning |
| title | Training self-learning circuits for power-efficient solutions |
| title_full | Training self-learning circuits for power-efficient solutions |
| title_fullStr | Training self-learning circuits for power-efficient solutions |
| title_full_unstemmed | Training self-learning circuits for power-efficient solutions |
| title_short | Training self-learning circuits for power-efficient solutions |
| title_sort | training self learning circuits for power efficient solutions |
| url | http://dx.doi.org/10.1063/5.0181382 |
| work_keys_str_mv | AT menachemstern trainingselflearningcircuitsforpowerefficientsolutions AT samdillavou trainingselflearningcircuitsforpowerefficientsolutions AT dineshjayaraman trainingselflearningcircuitsforpowerefficientsolutions AT douglasjdurian trainingselflearningcircuitsforpowerefficientsolutions AT andreajliu trainingselflearningcircuitsforpowerefficientsolutions |