重磅!发表在人工智能顶刊(TPAMI)上的一篇文章表示脑电与计算机视觉的交叉存在缺陷和盲点
更多技术干货第一时间送达
(左)图6来自Tirupattur等人的[7],这是据称由基于脑电波编码的GAN模型生成的样本图像(除了右边红色的列,它表示训练数据中给定类别的随机图像)。(右)几乎所有生成的图像都对应相同的ImageNet图像。请注意,左边据称合成的图像中,有一些(但不是全部)是右边ImageNet图像的水平镜像。还请注意,所有据称合成的图像都包含与相应的ImageNet图像相同的精确细粒度细节。特别是,每个图像不仅描绘了相应的类别,而且还描绘了与ImageNet对应的确切非类别特定的背景。
[1] C. Spampinato, S. Palazzo, I.Kavasidis, D. Giordano, N. Souly, and M. Shah, “Deep learning human mind forautomated visual classification,” in Proc. IEEE Conf. Comput. Vis. PatternRecognit., 2017, pp. 6809–6817.
[2] C. Spampinato, S. Palazzo, I.Kavasidis, D. Giordano, M. Shah, and N. Souly, “Deep learning human mind forautomated visual classification,” 2016, arXiv:1609.00344.
[3] S. Palazzo, C. Spampinato, I.Kavasidis, D. Giordano, and M. Shah, “Generative adversarial networksconditioned by brain signals,” in Proc. IEEE Int. Conf. Comput. Vis., 2017, pp.3410–3418.
[4] I. Kavasidis, S. Palazzo, C.Spampinato, D. Giordano, and M. Shah, “Brain2Image: Converting brain signalsinto images,” in Proc. 25th ACM Int. Conf. Multimedia, 2017, pp. 1809–1817.
[5] C. Du, C. Du, X. Xie, C. Zhang, and H.Wang, “Multi-view adversarially learned inference for cross-domain jointdistribution matching,” in Proc. Int. Conf. Knowl. Discov. Data Mining, 2018,pp.1348–1357.
[6] P. Kumar, R. Saini, P. P. Roy, P. K.Sahu, and D. P. Dogra, “Envisioned speech recognition using EEG sensors,” Pers. Ubiquitous Comput.,vol. 22, no. 1, pp. 185–199, 2018.
[7] P. Tirupattur, Y. S. Rawat, C.Spampinato, and M. Shah, “ThoughtViz: Visualizing human thoughts using generative adversarialnetwork,” in Proc. 26th ACM Int. Conf. Multimedia, 2018, pp. 950–958.
[8] S. Palazzo, C. Spampinato, I.Kavasidis, D. Giordano, and M. Shah, “Decoding brain representations bymultimodal learning of neural activity and visual features,” 2018,arXiv:1810.10974.
The Perils and Pitfalls of Block Design forEEG Classification Experiments
https://techxplore.com/news/2020-12-uncover-intersection-ai-neuroscience.html