Tero Keski-Valkama (MSc) is a technologist and a generalist. He is currently researching using deep neural networks for the purpose of anomaly detection in process traces in Flexible Assembly Systems, and large language model agent embodiment and meta-learning.

He has a long experience in artificial intelligence and machine learning.

Programming computers from early age, his interest in artificial intelligence and machine learning began when he was ten years old (in 1992) and followed instructions in a book for programming a simple inference engine for Commodore 64 BASIC capable of answering questions like: "IS CAT ANIMAL" ("YES", for previously inputted facts of "CAT IS MAMMAL" and "MAMMAL IS ANIMAL") (pg. 77).

In high school he experimented with simple perceptrons and Kohonen Self-Organizing Maps. He set up neural networks to learn how to play a simple game of Pong. He also made a simple reinforcement learning "pet" game where the pet could be conditioned to promote specific behavior.

In Tampere University of Technology he joined Roboteam and became the artificial intelligence and machine vision specialist. Taking part in several Eurobot competitions for autonomous mobile robotics he programmed planning and vision algorithms, optimizing different policies against each other in simulations. He also implemented an application which takes a photograph of a pile of decorative stones and separates the graphics as a hundred or so separate clipped images of single stones for use as graphical content in games and such. He also worked with data clustering and classification algorithms in the topic of signal processing. He has done everything from Sudoku-solvers to game AIs, from fuzzy logic to industrial process optimization.

Working in Elektrobit, he introduced a concept for car domain context dependent services based on semantic web.

Networking in social AI circles in different forums in the internet, he discovered Coursera and the leading edge knowledge about deep neural networks presented by the definitive people in the field: Geoffrey Hinton and Andrew Ng. To update his expertise he took all the other courses related to modern machine learning as well, including but not limited to deep neural networks, statistical analysis, regression, discrete optimization, computational neurobiology, GPU programming, and medical human brain physiology.

In the social networks he offered free consultation for people working with AI, and among others, he helped a friend in Russia in his research about synthetic emotions and emotional intelligence. He is also the lead administrator in the team administering all the official Transhumanist Party Facebook groups and pages in Europe.

He worked as a software architect in Cybercom mainly in web and industrial solutions, he took part in the first #Industryhack #HackTheFactory and his team won the competition with a machine learning concept with Markov Chains for anomaly detection. He also worked in the background group for most of the subsequent Industryhack competitions, and he helped creating the concept of social IoT forum (Cybercom Machinebook), and was in a central role in its development into a working concept later. He won his second #IndustryHack #HackTheShip competition in Germany in 2016.

He was the guild master in the Machine Learning guild in Cybercom Finland Oy, and the CTO in Sentido X-Tech Ltd. He was also one of the organizers of AI Morning.

He started researching practical applications for deep neural networks and other modern machine learning technologies aiming towards a doctoral degree. The topic of research is applying deep neural networks for interleaved process traces from Flexible Assembly Systems. He has also countless other open source and hobby projects related to the field, and he actively follows the latest developments in relevant technologies.

Currently he is working as a staff software engineer in Alloy.ai.