Public Engagement with Science?
First of all, I know of nobody who is not part of the public. The issue is, it is unfortunate for the public that is not doing science and research at University or in their free time as a hobby, to miss out on the beauty that nature unfolds almost every moment in time. Of course, to begin to understand each moment, it often requires centuries of work. But, this work culminates in knowledge that can only inspire if told in an accessible manner.
I believe in making science accessible. If you are someone who would like to have something understood just in person, or have something understood by an audience (at a cafe or pub or lecture theatre) of whatever size you will, feel free to contact me. I do this for free, just in case you wondered, and preferably after 1700hrs on weekdays. If the topic that you are interested in is anywhere within the realms of Computer Science (Artificial Intelligence in particular), Mathematics, Economics or Physics, I would require less time (at least two weeks however) to prepare. Otherwise, please give me a month’s notice.
An updated (Sept. 2016) lecture on reinforcement learning, covering more concepts than my previous lectures on the subject (see below).
How can machines go about making decisions in the face of uncertainty, towards achieving various long term goals? How can they build on their experiences for such decision making? How can they know when it might be the right time to go whacky and make risky decisions, and when not to do so? This lecture discussed these questions within a framework that has come to be known as reinforcement learning. Machines of today are increasingly becoming autonomous entities, as if playing games along with, against or on behalf of us humans. They learn with every move, adapt their behaviour and strive to gain a better understanding of the problems they are designed to tackle. At least that's how machines will be described by us in the not so very distant future. Recent advances in reinforcement learning using deep architectures point in that direction as well. From playing board games, to getting rovers to work on Mars, to more interesting health and education issues that necessitate computing devices to continuously process data, reinforcement learning forms a fundamental piece of such puzzles.
This was an introductory lecture mildly covering one of the branches of machine learning where there is much data but one does not know of what practical use it might be. One can try to analyse it, visualise it - just try to make sense of it. I usually like to get my students play games. So, we had the whole class come on 'stage', hold hands, and move about in the classroom as I threw stuff on the floor. Well, that's how self-organising maps work, don't they? It is always nice to enact the simple rules that lead to beautiful emergent behaviour, even if one ends up reducing each student to a single 'neuron' in the process.
This lecture focussed on the need for redundancy in messages for robust long distance communication. The idea was to show how redundancy (extra drum beats/symbols/bits) in sent messages, whether communicated via drumming centuries ago in Africa, or via mobile phones today, helps with recovering information from it after having been corrupted in transit. Many high school students from two schools in Oslo, Elvebakken and Oslo Katedralskole were present. Two of them got to play drums on the stage in our very own redundant orchestra, while I got to play the conductor.
Event: Realfag spesial - populærvitenskapelige kortforedrag