The n00best path to data science and machine learning state of the art is now complete, no excuses! 2015: Andrew Ng’s Coursera MOOC; 2016: Kaggle competitions with xgboost and ensembles; 2017: deep learning code-oriented courses with fast.ai and GPU hardware for the masses, part 2 just gone public today. Thanks, very lucky to witness and try this.
Molten Salt nuclear reactors are being developed at commercial startup level and through the institutional Gen IV International Forum. In particular, the EU is now working at SAMOFAR , its Molten Salt fast reactor, within the Horizon 2020 framework under the leadership of Prof. Elsa Merle-Lucotte from Grenoble University, France. The presentations from their recent Summer School are now online, they add nicely to my database. Molten Salt reactors are the less technologically ready among the Gen IV designs and, as such, still open to independent contributions: layouts, critical components and materials from me?
Italian poetry is locked into its 1960s loop of elitism, moral superiority and pre-massmediatic imagery. Therefore, nothing can be shared with the people running that art, they are entombed. The forefront of aesthetics for the public moved towards visual arts in the 1990s thanks to collective exhibitions and, in my opinion, will meet thematic hackathons in the 2020s. The basic principles remain the same, it is the form that changes, aggregation and insubordination will persist. I do not think I will have a part in that, I am more and more convinced that art is a private matter living through a personal journey.
My first image classification, deep learning competition on Kaggle ended last night. It was Kaggle Planet and I finished 172/938 , unimpressive but pretty solid for a noob. It was a multilabel classification problem with 100k images from the Amazon Forest, for which I implemented a few state of the art architectures in Keras and ensembled their results. Help came from public kernels and forum topics, so a great learning experience on top of the fast.ai MOOC by Jeremy Howard and the latest Deep Learning with Python book by Francois Chollet. Now putting all together for a post-competition Jupyter notebook I can reuse for similar tasks in the future. Thanks, Kaggle!
Molten Salt reactors are the most exotic Gen IV nuclear power plant concept and still undemonstrated. This book Molten Salt Reactors and Thorium Energy from Prof T.J. Dolan and Elsevier introduces their state of the art at June 2017. One of the most interesting startup within this industry is Moltex Energy from the UK, whose Stable Salt Reactor concept appears to be relying upon technologically ready materials. They are working actively with Canada for a possible demonstration there in the 2020s.
I am becoming pretty used to Keras for image classification deep learning but still need control of the size and the time required to train and test my networks. One procedure I absolutely need to master in a short time is fit_generator() , which makes training / testing computations batch by batch instead of storing all the images in memory, 16GB RAM going away pretty fast. This would help me process very big datasets and bigger image sizes than the 64×64 pixels I am using now. DenseNet-121 is ok with several tricks and 5 k-folds already, but ResNet-50 and Inception-4 are not yet, both requiring 224×224 images.
Last week the Nuclear Engineering International magazine published a contribution about additive manufacturing for nuclear power. Among industrial and research case studies, it is explained that: “There are a number of applications for additive manufacturing in the nuclear sector, including in new-build, for fuel and for in-reactor components. However, work still needs to be done to qualify material and demonstrate that components can meet nuclear codes and standards.” As a PhD holder in Materials Science for nuclear energy, I am surely watching with both eyes wide open and taking note for my own machine learning research.
“The International Conference on Fast Reactors and Related Fuel Cycles: Next Generation Nuclear Systems for Sustainable Development (FR17), hosted by the Government of the Russian Federation through State Atomic Energy Corporation ROSATOM, provides a forum to exchange information on national and international programmes as well as new developments and experiences. The 26-29 June gathering in Yekaterinburg is the third such international conference after previous editions in Kyoto, Japan (2009), and Paris, France (2013).” cit. IAEA <=> 450+ state of the art new potential contributions to my nuclear papers database for machine learning research. Thanks!
I got in contact with a very interesting applied research business operating in the materials science field between Rotterdam and Bologna. Parx Plastics produces antibacterial plastics and is developing a new line of biocompatible antimicrobial fabrics, for which a transfer learning from a senior researcher is planned soon. They are pretty well introduced and also trying to disseminate their results. My aim at the moment is to familiarise with their methods, possibly reporting back to them in a couple of weeks if suited.
I use to spend one or two months in Turin every year both as a tourist and to keep in touch with professional fellows from the North-West Italy, while looking for gig opportunities. It seems a bit more difficult this year because of ongoing commitments in Foggia. October may be the time but cannot say yet, therefore cannot book my stay. We will see.