I’m currently working at the National Institute of Scientific Research (INRS) in Canada, where I am developing applied solutions in two different fields: Machine learning and Energy harvesting.
In the first topic, Artificial Neural Network are the main focus of interest. Much like the brain, these software algorithms are trained via repetitions, with each iteration reinforcing specific neural pathways until an approximately suitable solution can be found. With the blooming of computer-assisted decision making, these programs are now at the heart of a wide variety of fields; search engines, voice recognition, vehicular/robotic automation and financial trading systems. This iterative learning process also lies at the root of their limitation in size and speed, since the state of each neuron and connection has to be calculated with respect to each other. This updating process therefore makes the system exponentially complex when increasing the number of neurons.
I have recently invented a neural network architecture, in which the state of neurons and connections are not calculated, but rather the result of physical laws. By avoiding the problem of calculating the state of each neurons, the regression imposed by the learning process becomes a one-step operation, independently of the size of the neural network. This way, we can unlock the full potential of artificial neural networks and offer significant improvements in the field of machine learning.
In all countries, the vast majority of produced energy is wasted through thermal loss: In the USA and China, for example, over 90% of power generation comes from thermal processes (coal, nuclear, natural gas), having 30-60% of conversion efficiency. This represents billions of tonnes of oil equivalent (Gtoe) in dissipated energy and gigatonnes (Gt) of additional CO2 emissions. In the oil refining and metal making industries, 20-50% of the energy consumed during the manufacturing process is lost via waste heat. A cost-effective way of generating energy from heat, for applications in both energy harvesting and waste recovery, would have vastly positive impacts in fields of energy management and in the reduction of industrial greenhouse gases.
My second topic takes advantage of the same physical phenomena used in my physical neural network to offer a thermionic power conversion mechanism for the recycling of wasted heat. A direct conversion of heat into electrical energy is achieved through a combination of Seebeck effect and self-assembled nanostructures capable of yielding high voltages for small temperature gradients. I aim to develop this to into a roll-to-roll thin film technology to produce a cost-effective foil capable of harvesting electric energy when in contact with heated objects.