Developed by Padhu University, the University of Padhu, etc. to enable on -demand re -programming IC chips and continuous learning of AI
delivery
3コメント3件IC chip developed by the research team
米パデュー大学と中国の北京航空航天大学、米ペンシルベニア州立大学、米サンタクララ大学、米Argonne National Laboratory、米University of Illinois Chicago、米Brookhaven National Laboratory、米ジョージア大学、米ポートランド州立大学による研究チームがThe developed "Reconfigurable Perovskite Nickelate Electronics for Artificial Intelligence" is an on -demand IC chip. Supports neural networks continuously learning like a human brain. Stirmingly inputting new data into a neural network will interfere with the knowledge you have learned earlier and reduce performance. Forgetting past learning in this way is called catastrophic forgetting, which remains as one of the issues. On the other hand, human cranial nerves have a plasticity of plasticity that constantly perform structural changes due to external stimuli, etc., so they form a new connection between neurons and adapt to new information continuously. In this way, the brain always changes as needed, but since the IC chip circuit does not change, continuous learning by computer has been extremely difficult.This indicates the need to make the hardware dynamic, not the software side. In the future, it will be important in configuring machines equipped with highly autonomous machine learning. Research proposes a device for the IC chips dynamically re -programs to incorporate new data like the brain, and to help computers continue to learn over time.This device is composed of nickel oxide (Renio3) with a pero -scity structure. Renio3 has the property of changing the electrical conductivity significantly due to a phase transfer from a metal to an insulator (or an insulator to a metal). Renio3 is also known to significantly increase electric resistance at room temperature by dope (adding impurities to change physical properties) of light and small small proton (hydrogen ions). Furthermore, pre -research has been successful in controlling the hydrogen ion. Proton dope's resistance modulation is greater than before. Devices that combine proton dope and Renio3 are attracting attention because normal electronic control can be performed.This device was applied to IC chips. As a result of simulating deep -struck learning frameworks in the experimental data, networks using this device are more excellent in tasks such as numerical recognition and ECG pattern classification than static networks. rice field. In addition, it was confirmed that it was stable even after repeating the state switching of more than 1 million cycles. Source and Image Credits: Zhang, H.T.et al.“Reconfigurable Perovskite Nickelate Electronics for Artificial Intelligence” Science.3 Feb 2022, Vol 375, ISSUE 6580, PP.533-539 Doi: 10.1126/Science.ABJ7943 * Mr. Yuki Yamashita, who presides over the web media "SEAMLESS", introducing the latest research on technology.Mr. Yamashita picks up and explains a highly new nature paper.
ITMEDIA NEWS
最終更新:ITMEDIA NEWS
