本研究建構了一個腦信號輔助的人工智慧計算系統。我們將腦波的信
號加入人工智慧類神經網路的參數中，進行訓練與計算。我們比較加入腦波
資料前後的信號辨識能力。
人工智慧系統的輸入信號為三角波與方波信號，我們加入雜訊以增加
辨識困難度，相對應於三角波與方波信號的輸出信號為0與1。我們讓受試者
聆聽此三角波與方波信號，同時以腦波儀擷取受試者的腦波。我們使用
Reservoir computing（RC）[1]的計算技術，將腦波訊號以短時間傅立葉轉換
處理後，將不同頻率（腦波的信號範圍為 0 到 70 Hz的頻率）的腦波信號
合併輸入信號，一起輸入人工智慧系統進行計算。
在RC計算技術中，需以亂數定義輸入權重矩陣（ 𝑊𝑖𝑛 ）與遞迴權重矩
陣（ 𝑊 ）。由本實驗可得3個結果： 1. 固定腦波資料，我們隨機產生 𝑊𝑖𝑛
與 𝑊 矩陣數次進行計算，發現對計算結果的影響程度小。 2. 在某一位受
試者上，於不同時間點所量測的 10 次結果，可以發現，在 8、12、52、54
與 68 Hz 所計算出來的 10 次均方誤差 MSE 的變異性較大，具有因時而
異的特性。 3. 以14-42 Hz 與 58-62 Hz 的腦波信號加入 RC 計算程式中，
結果發現，11位受試者的結果，均能有效降低輸入信號的辨識錯誤率。此結
果可結論出，以腦波加入人工智慧計算系統，比沒有加入腦波的人工智慧計
ii
算系統，更能精確辨識輸入信號。也就是以腦波輸入人工智慧系統，更能精
確地提供適合每個人需要的結果。
本研究以腦波加入人工智慧的計算當中，結果（使用14-42 Hz 與 58-
62 Hz腦波信號）可有效降低信號的辨識錯誤率。此降低的程度，與每個人
在受測當時的狀況有關。此成果（使用8、12、52、54 與 68 Hz腦波信號）
也展示了因時因人因環境不同所產生的反應（個性與喜好、或當下的情緒、
或生理心理的狀況）。未來可延伸應用至判斷個人的喜好，或將逝者的人腦
的反應信號加入計算當中，可讓人類的個性與意見長久保留。
本研究計畫經由國立臺灣大學行為與社會科學研究倫理委員會審核通
過，倫委會案號: 202103EM011（見附錄一）。

Artificial intelligence computation system assisted by brainwave signal has been
built in this study. We add the electroencephalography as input data to the neural
networks to optimize the computing system. We examine the signal recognition
ability before and after adding brainwave data.
The input signals consist of the triangular and rectangular waves. The waves are
incorporated with white Gaussian noise to increase the difficulty of signal
recognition. The corresponding output signals to the input triangular and
rectangular waves are 0 and 1, respectively. The subjects listen to the triangular
and rectangular waves. The brainwave data of the subjects are collected by using
MindWave mobile. Using the short-time Fourier transform (STFT) with a
nonoverlapped Hamming window, the brainwaves are transformed to
encephalogram(EEG) spectra. Using Reservoir Computing (RC), the EEG signals
of different frequencies were added into the RC program as input data.
In RC. 𝑊𝑖𝑛 and 𝑊 are the input and recurrent weight matrices, respectively,
which consist of random numbers. We obtain three main results:
1. The value in the arbitrarily generated 𝑊𝑖𝑛 and 𝑊 matrices have small
influence on the calculation results.
2. In the measured results for 10 different times on the same subject, it can be
found that brainwave calculations at 8, 12, 52, 54 and 68 Hz, the MSE results have
significant difference. They have characteristics that vary for different time.
3. With the EEG signal from 14 to 42 Hz and from 58 to 62 Hz frequency, the
signal recognition error rate can be reduced effectively.
iv
Brainwave signals were added into artificial intelligence computation system in
this study. The signal recognition error rate can be reduced effectively (using the
EEG signals at 14-42 Hz or 58-62 Hz). This result (using the EEG signals at 8,
12, 52, 54 and 68 Hz) also demonstrates the artificial intelligence computation
involved in different human responses (like personality and preferences, emotions
or physical and psychological conditions). In the future, it can be extended to
determine personal preferences. If one day the brain of the deceased persons could
be kept active, the response signal of the brain can also added into the computation.
It might make the personality or the spirit immortal.
This research project was approved by Research Ethics Committee National
Taiwan University. NTU-REC No.：202103EM011. (See Appendix 1)

1. M. Lukoševičius, H. Jaeger, and B. Schrauwen, KI - Künstliche Intelligenz 26, 365-371
(2012).
2. W. S. Bainbridge, Science 261, 1186-1187 (1993).
3. T. Chouard, Nature (2016).
4. D. E. Michel, The American Music Teacher 27, 18 (1977).
5. E. A. Feigenbaum, State of the art report on machine intelligence. Maidenhead:
Pergamon-Infotech (1981).
6. G. E. Hinton, and R. R. Salakhutdinov, science 313, 504-507 (2006).
7. G. E. Hinton, S. Osindero, and Y.-W. Teh, Neural computation 18, 1527-1554 (2006).
8. H. Jaeger, Bonn, Germany: German National Research Center for Information
Technology GMD Technical Report 148, 13 (2001).
9. M. Lukoševičius, and H. Jaeger, Computer Science Review 3, 127-149 (2009).
10. H. Jaeger, "Reservoir riddles: Suggestions for echo state network research," in
Proceedings. 2005 IEEE International Joint Conference on Neural Networks,
2005.(IEEE2005), pp. 1460-1462.
11. P. Gloor, American Journal of EEG Technology 9, 1-8 (1969).
12. H. Berger, Archiv für Psychiatrie und Nervenkrankheiten 87, 527-570 (1929).
13. F. Rauscher, G. Shaw, L. Levine, E. Wright, W. Dennis, and R. Newcomb, Neurological
research 19, 2-8 (1997).
14. J. Bhattacharya, and H. Petsche, Proceedings of the Royal Society of London. Series B:
Biological Sciences 268, 2423-2433 (2001).
15. K. Natarajan, R. Acharya, F. Alias, T. Tiboleng, and S. K. Puthusserypady, Biomedical
engineering online 3, 1-11 (2004).
16. E. R. Miranda, and A. Brouse, Leonardo 38, 331-336 (2005).
17. Y.-P. Lin, C.-H. Wang, T.-P. Jung, T.-L. Wu, S.-K. Jeng, J.-R. Duann, and J.-H. Chen,
IEEE Transactions on Biomedical Engineering 57, 1798-1806 (2010).
18. M. W. K. M. Dr. Aziz Makandar, Journal of Information and Computational Science 10
(2020).
19. J. Herbert, "Adaptive nonlinear system identification with echo state networks," in
Proceedings of the 15th International Conference on Neural Information Processing
Systems(MIT Press, 2002), pp. 609–616.
20. W. Maass, T. Natschläger, and H. Markram, Neural computation 14, 2531-2560 (2002).
21. W. Maass, Computability in context: computation and logic in the real world, 275-296
(2011).
22. J. Dai, G. K. Venayagamoorthy, and R. G. Harley, "An introduction to the echo state
network and its applications in power system," presented at the 2009 15th International
58
Conference on Intelligent System Applications to Power Systems2009.
23. M. Lukoševičius, "A practical guide to applying echo state networks," in Neural
networks: Tricks of the trade(Springer, 2012), pp. 659-686.
24. A. Ben-Israel, and T. N. Greville, Generalized inverses: Theory and applications
(Springer Science & Business Media, 2003).
25. A. H. Murphy, Monthly Weather Review 116, 2417-2424 (1988).
26. P. A. Abhang, B. W. Gawali, and S. C. Mehrotra, "Chapter 2 - technological basics of
eeg recording and operation of apparatus," in Introduction to eeg- and speech-based emotion
recognition, P. A. Abhang, B. W. Gawali, and S. C. Mehrotra, eds. (Academic Press, 2016),
pp. 19-50.
27. H. Marzbani, H. R. Marateb, and M. Mansourian, Basic and clinical neuroscience 7, 143-
158 (2016).
28. J. Fell, G. Fernandez, P. Klaver, C. E. Elger, and P. Fries, Brain Research Reviews 42,
265-272 (2003).
29. A. W. Keizer, M. Verschoor, R. S. Verment, and B. Hommel, International Journal of
Psychophysiology 75, 25-32 (2010).
30. A. W. Keizer, R. S. Verment, and B. Hommel, Neuroimage 49, 3404-3413 (2010).
31. S. Staufenbiel, A.-M. Brouwer, A. Keizer, and N. Van Wouwe, Biological psychology
95, 74-85 (2014).
32. T. Egner, and J. H. Gruzelier, Neuroreport 12, 4155-4159 (2001).
33. D. Vernon, T. Egner, N. Cooper, T. Compton, C. Neilands, A. Sheri, and J. Gruzelier,
International journal of psychophysiology 47, 75-85 (2003).
34. T. Egner, and J. H. Gruzelier, Clinical neurophysiology 115, 131-139 (2004).
35. W. Klimesch, M. Doppelmayr, H. Russegger, T. Pachinger, and J. Schwaiger,
Neuroscience letters 244, 73-76 (1998).
36. W. Klimesch, H. Schimke, and J. Schwaiger, Electroencephalography and clinical
Neurophysiology 91, 428-441 (1994).
37. D. J. Vernon, Applied psychophysiology and biofeedback 30, 347-364 (2005).
38. T. Dempster, An investigation into the optimum training paradigm for alpha
electroencephalographic biofeedback (Canterbury Christ Church University (United
Kingdom), 2012).
39. W. Klimesch, M. Doppelmayr, H. Schimke, and B. Ripper, Psychophysiology 34, 169-
176 (1997).
40. D. B. Jarrett, J. B. Greenhouse, J. M. Miewald, I. B. Fedorka, and D. J. Kupfer,
Biological psychiatry 27, 497-509 (1990).
41. T. Sürmeli, and A. Ertem, Journal of Neurotherapy 11, 63-68 (2007).
42. C. J. Davis, J. M. Clinton, K. A. Jewett, M. R. Zielinski, and J. M. Krueger, Journal of
clinical sleep medicine 7, S16-S18 (2011).
43. J. N. Demos, Getting started with neurofeedback (WW Norton & Company, 2005).
59
44. J. R. Evans, and A. Abarbanel, Introduction to quantitative eeg and neurofeedback
(Elsevier, 1999).
45. D. Newandee, "Measurement of the electroencephalogram (eeg) coherence, atmospheric
noise, and schumann resonances in group meditation," (1996).
46. Y. Cui, and Y. Wang, TVBA-5000 (2018).
47. H. J. Nussbaumer, "The fast fourier transform," in Fast fourier transform and
convolution algorithms(Springer, 1981), pp. 80-111.
48. R. N. Bracewell, and R. N. Bracewell, The fourier transform and its applications
(McGraw-Hill New York, 1986).
49. J. B. Allen, and L. R. Rabiner, Proceedings of the IEEE 65, 1558-1564 (1977).
50. M. Portnoff, IEEE Transactions on Acoustics, Speech, and Signal Processing 28, 55-69
(1980).
51. F. J. Harris, Proceedings of the IEEE 66, 51-83 (1978).
52. I. NeuroSky, "What can you do with mindwave mobile 2?,"
https://store.neurosky.com/pages/mindwave.
53. W. Sałabun, Przeglad Elektrotechniczny 90, 169-174 (2014).