Think Outside the Box! [From the Editor]

Summer is a nice, quiet period, with the end of the academic year in many countries. It is a time for vacations and to think about events and develop a vision for our future. Of course, this vision is based on current research results, biased by fashion effects and publish or perish tendencies and perturbated by the COVID-19 pandemic, which, in depth, modifies our life—professional as well as personal. First, I would like to evoke the memory of Prof. Jeanny Hérault, who died on 7 June 2021. Since 1970, Jeanny had been a faculty member at the Polytechnic Institute and then at the University Joseph Fourier in Grenoble, France, and he trained dozens of Ph.D. students, postdocs, and young researchers. He was my Ph.D. supervisor before becoming my friend, and I am sure that most of you understand from your own experiences how strong our friendships could be. Some of you knew him as well as his contributions: he was actually an out-of-the-box scientist. In this editorial, I will recall some of his outstanding contributions around signal processing. Very early on, Jeanny was fascinated by the brain, as it is a fantastic machine for signal processing. When he decided to prepare a Ph.D. in 1968, he chose a topic fully out of the box focused on modeling neural cells and simulating them on electronic circuits. And then, during his entire career, he proposed innovative methods in signal and image processing always inspired by brain and human vision, including artificial neural networks. In early 1980, inspired by modeling how the brain of vertebrates can decode arm and leg motions, he proposed the basics of blind source separation and independent component analysis [1]. In the 1990s, since computers were not powerful enough to run artificial neural networks, Carver Mead [2] in the United States and a few scientists around the world developed, like Jeanny and several of his Ph.D. students, a few special architectures able to train and efficiently use neural networks such as multilayer perceptrons and self-organizing maps [3]. He then developed the concept of curvilinear component analysis, which extends the Kohonen’s maps to huge-dimension data and implicitly considers high-dimension data are embedded in a low-dimension manifold with suited non-Euclidian metrics [4]. Finally, in the mid-1990s, he focused on modeling retina, first the neural layers after the retina and more generally vertebrate visual perception, for understanding how motions, textures, colors, etc., are processed in the brain. These works result in many powerful bioinspired algorithms, as well as applications and patents, which are summarized in his very inspiring book Vision: Images, Signals and Neural Networks. Models of Neural Processing in Visual Perception [5] published in 2010. Finally, just after he retired, he created a nonprofit association in Grenoble, open to any curious person, called Neurocercle (https:// neurocercle.wordpress.com), and each month since 2009, he has invited a scientist to talk about advances in neurosciences and cognitive sciences and implications in our lives.