3 ms·

Ultra-efficient machine learning transistor cuts AI energy use by 99%

16 points by westurner 3y ago

westurner 3y ago
"Reconfigurable mixed-kernel heterojunction transistors for personalized support vector machine classification" (2023) https://www.nature.com/articles/s41928-023-01042-7 https://www.nature.com/articles/s41928-023-01042-7 : > Advances in algorithms and low-power computing hardware imply that machine learning is of potential use in off-grid medical data classification and diagnosis applications such as electrocardiogram interpretation. However, although support vector machine algorithms for electrocardiogram classification show high classification accuracy, hardware implementations for edge applications are impractical due to the complexity and substantial power consumption needed for kernel optimization when using conventional complementary metal–oxide–semiconductor circuits. Here we report reconfigurable mixed-kernel transistors based on dual-gated van der Waals heterojunctions that can generate fully tunable individual and mixed Gaussian and sigmoid functions for analogue support vector machine kernel applications. We show that the heterojunction-generated kernels can be used for arrhythmia detection from electrocardiogram signals with high classification accuracy compared with standard radial basis function kernels. The reconfigurable nature of mixed-kernel heterojunction transistors also allows for personalized detection using Bayesian optimization. A single mixed-kernel heterojunction device can generate the equivalent transfer function of a complementary metal–oxide–semiconductor circuit comprising dozens of transistors and thus provides a low-power approach for support vector machine classification applications.
- westurner 3y ago
  Heterojunction: https://en.wikipedia.org/wiki/Heterojunction https://en.wikipedia.org/wiki/Heterojunction dual-gated van der Waals heterojunctions: Anyways, the Heterojunctions article mentions quantum dots, which can be fabricated with [synthetic] DNA: https://news.ycombinator.com/item?id=37698081 https://news.ycombinator.com/item?id=37698081 : > "Ultrafast dense DNA functionalization of quantum dots and rods for scalable 2D array fabrication with nanoscale precision" (2023) https://www.science.org/doi/10.1126/sciadv.adh8508 https://www.science.org/doi/10.1126/sciadv.adh8508
westurner 3y ago
On separable classes, separable states, SVMs without random initialization, and quantum annealing SVMs on QC: https://news.ycombinator.com/item?id=37367951#37378940 https://news.ycombinator.com/item?id=37367951#37378940 SVM: https://en.wikipedia.org/wiki/Support_vector_machine https://en.wikipedia.org/wiki/Support_vector_machine

16 points by westurner 3y ago

"Reconfigurable mixed-kernel heterojunction transistors for personalized support vector machine classification" (2023) https://www.nature.com/articles/s41928-023-01042-7 https://www.nature.com/articles/s41928-023-01042-7 : > Advances in algorithms and low-power computing hardware imply that machine learning is of potential use in off-grid medical data classification and diagnosis applications such as electrocardiogram interpretation. However, although support vector machine algorithms for electrocardiogram classification show high classification accuracy, hardware implementations for edge applications are impractical due to the complexity and substantial power consumption needed for kernel optimization when using conventional complementary metal–oxide–semiconductor circuits. Here we report reconfigurable mixed-kernel transistors based on dual-gated van der Waals heterojunctions that can generate fully tunable individual and mixed Gaussian and sigmoid functions for analogue support vector machine kernel applications. We show that the heterojunction-generated kernels can be used for arrhythmia detection from electrocardiogram signals with high classification accuracy compared with standard radial basis function kernels. The reconfigurable nature of mixed-kernel heterojunction transistors also allows for personalized detection using Bayesian optimization. A single mixed-kernel heterojunction device can generate the equivalent transfer function of a complementary metal–oxide–semiconductor circuit comprising dozens of transistors and thus provides a low-power approach for support vector machine classification applications.

westurner 3y ago
Heterojunction: https://en.wikipedia.org/wiki/Heterojunction https://en.wikipedia.org/wiki/Heterojunction dual-gated van der Waals heterojunctions: Anyways, the Heterojunctions article mentions quantum dots, which can be fabricated with [synthetic] DNA: https://news.ycombinator.com/item?id=37698081 https://news.ycombinator.com/item?id=37698081 : > "Ultrafast dense DNA functionalization of quantum dots and rods for scalable 2D array fabrication with nanoscale precision" (2023) https://www.science.org/doi/10.1126/sciadv.adh8508 https://www.science.org/doi/10.1126/sciadv.adh8508

westurner 3y ago

On separable classes, separable states, SVMs without random initialization, and quantum annealing SVMs on QC: https://news.ycombinator.com/item?id=37367951#37378940 https://news.ycombinator.com/item?id=37367951#37378940 SVM: https://en.wikipedia.org/wiki/Support_vector_machine https://en.wikipedia.org/wiki/Support_vector_machine