Difference between revisions of "Team:SEU/Description"
| Line 135: | Line 135: | ||
<p>[3] Cherry, Kevin M., and Lulu Qian. "Scaling up molecular pattern recognition with DNA-based winner-take-all neural networks." Nature 559.7714 (2018): 370.</p> | <p>[3] Cherry, Kevin M., and Lulu Qian. "Scaling up molecular pattern recognition with DNA-based winner-take-all neural networks." Nature 559.7714 (2018): 370.</p> | ||
<p></p> | <p></p> | ||
| − | <p>[4] <a href="https://www.youtube.com/watch?v=zjC_0PW9A4c&feature=youtu.be&tdsourcetag=s_pctim_aiomsg">https://www.youtube.com/watch?v=zjC_0PW9A4c&feature=youtu.be&tdsourcetag=s_pctim_aiomsg</a> | + | <p>[4] <a href="https://www.youtube.com/watch?v=zjC_0PW9A4c&feature=youtu.be&tdsourcetag=s_pctim_aiomsg">https://www.youtube.com/watch?v=zjC_0PW9A4c&feature=youtu.be&tdsourcetag=s_pctim_aiomsg</a></p> |
<p></p> | <p></p> | ||
| − | <p>[5] <a href="https://www.youtube.com/watch?v=42FCzoJt8Pg&feature=youtu.be">https://www.youtube.com/watch?v=42FCzoJt8Pg&feature=youtu.be</a> | + | <p>[5] <a href="https://www.youtube.com/watch?v=42FCzoJt8Pg&feature=youtu.be">https://www.youtube.com/watch?v=42FCzoJt8Pg&feature=youtu.be</a></p> |
<p></p> | <p></p> | ||
</div> | </div> | ||
Revision as of 08:21, 1 October 2019
Description
Artificial intelligence is one prevailing research field in recent years due to its successful application in areas like medical care, computer vision and natural language processing. Most of artificial intelligence systems are implemented with silicon-based chips which have high area complexity and power cost when facing large systems. Is it possible to use bio-chemical materials to implement artificial intelligence systems more efficiently?
In this project, we have employed DNA reactions to implement neural networks which are basic components in artificial intelligence systems. Specifically, using molecular concentrations as input and output signals, we propose and demonstrate computation models for weighted summation (including multiplication and addition) and activation functions (e.g., ReLU and Sigmoid) which are arithmetic operations in neural networks. Also, the classical backpropagation algorithm which is utilized in the training process can be similarly constructed (which has not been studied in molecular computing to our best knowledge). Based on these models, we synthesize basic neural networks with DNA strand displacement reactions.
To help other researchers obtain the proposed DNA-based artificial neural networks, we develop a software tool to automatically generate DNA reaction models and relevant DNA sequences according to the required parameters of neural networks, which may help them with further bio-computer design and bio-robot design.
References
[1] C. Fang, Z. Shen, Z. Zhang, X. You and C. Zhang, "Synthesizing a Neuron Using Chemical Reactions," IEEE International Workshop on Signal Processing Systems (SiPS), Cape Town, 2018, pp. 187-192.
[2] Wilhelm, Daniel, Jehoshua Bruck, and Lulu Qian. "Probabilistic switching circuits in DNA." Proceedings of the National Academy of Sciences 115.5 (2018): 903-908.
[3] Cherry, Kevin M., and Lulu Qian. "Scaling up molecular pattern recognition with DNA-based winner-take-all neural networks." Nature 559.7714 (2018): 370.
[4] https://www.youtube.com/watch?v=zjC_0PW9A4c&feature=youtu.be&tdsourcetag=s_pctim_aiomsg
[5] https://www.youtube.com/watch?v=42FCzoJt8Pg&feature=youtu.be
