For those unaware, the Dunning-Kruger is a cognitive bias where people with low ability at a task tend to overestimate their ability. The cause of this is usually being unaware of the expected standards for a task. The field of data science and machine learning has attracted a lot of attention in the past several years, regularly appearing in the top 10 jobs of the year for salary, prospects, and job satisfaction. These attractive statistics have caused an influx of people to pivot in their careers.
For those pivoting to a career in data science and who do not have the time or the means to attend a higher education institution, a common path to take is enrolling in a Massive Online Open Course (MOOC) on platforms such as Coursera, Udemy, or Udacity. Some of the courses that usually get recommended are:
- Andrew Ng Machine Learning — Coursera
- Deep Learning Specialisation — Coursera
- Introduction to machine learning for coders — Fast.ai
I am a fan of these courses. In the early stages of trying to forge a career in machine learning, they introduce the building blocks well. The issue with these courses is that they are designed to be accessible to the layman, those with no experience or expertise. In catering to this group the courses lack the depth of knowledge needed to become an expert. However, on the completion of one of these courses, you may feel like a machine learning aficionado, able to solve the world’s problems with your multi-layer perceptron. It is this point where people are at the peak of the Mountain of Stupidity (Figure 1). It may take a non-trivial task or encountering a colleague with a much better understanding and experience in the field to bring us down from this peak and realize our own failings.
We not only realize the depth of knowledge we lack but also the breadth. Imposter syndrome may set in at this stage. This is where you begin to doubt your own ability and undervalue the previous successes that have gotten you to the position you are currently in. This is natural and can even be felt by people who we would deem experts. Upon reaching this stage the next natural step is to find resources that will banish this feeling of being an imposter.
Machine learning is a broad term used to cover fields that an entire lifetime could be spent researching: Natural Language Processing, Computer Vision, Reinforcement Learning, Optimisation, etc. Are we expected to buy a textbook for every field? What about the sub-fields? Natural Language Processing can be broken up into many sub-fields: Machine Translation, Named Entity Extraction, Natural Language Generation, Automatic Speech Recognition, the list goes on. A dedicated textbook can be written for each.
The other drawbacks to textbooks are:
- They are expensive and not everyone is in the position to spend a hundred dollars on a textbook
- Breakthroughs and innovation move so quickly that any textbook that you buy is likely to be outdated in a matter of months to a couple of years.
- Textbooks that cover multiple fields in machine learning have the same drawbacks as the MOOC courses, they offer a too general an insight rather than getting to the nuts and bolts of a topic.
How then to attain knowledge without breaking the bank for it?
To combat these issues of cost, relevancy, and expertise, I would like to introduce SlidesLive. SlidesLive is a professional conference recording platform that allows for participants in conferences to perform their presentations virtually.
It cannot be said that many positives have been brought by Covid-19 but one of them is the potential democratization of academic research. Current restrictions on travel have caused many conferences to move to a completely online format. This brings itself a whole host of issues to tackle: How will people present their paper effectively? How can an international conference accommodate different time zones across the globe? A popular solution appears to be SlidesLive. The latest breakthroughs that previously required a conference attendance fee to access are now available for free in the SlidesLive Library.
Conferences are able to host the presentations of papers and keynotes on the SlidesLive Library. I have linked a keynote that was presented at ICML — International Conference of Machine Learning as an example.
The layout demonstrated in this talk is the standard for presentations hosted on the platform. The left-hand side of the screen is for the presenter to appear and the right-hand side is for the slides that accompany the presentation. One of the most useful features is the ability to sync the slides to the presentation. If there is a concept you didn’t quite understand you can sync the recording of the presentation to the slide that introduced the concept and listen to the presentation again. On the other hand, if you want to skip the introductory parts of a presentation you can go to the slides you want to start the recording from. No messy rewinding and fast-forwarding.
It should be noted at this stage that not all the conferences on SlidesLive are related to machine learning. There are a plethora of conferences out there and SlidesLive caters to all. An example of some of the other content available:
- The Power of Community
- How to be an android expert
- The Architecture of Understanding
- The opportunity to learn about cutting edge research from experts and gain insight into the future direction of the field by leading voices.
- The availability and cost to access the research and having the latest papers in your field of interest organized for you in consumable chunks is a nice bonus.
- The ability to sync the slides to the recording is an efficient method of consuming the presentation.
- The democratization of research is not perpetrated by every conference that uses SlidesLive. NeurIPS has over 4500 presentations on SlidesLive however none of them have been listed and you will only be able to access them if you attended the conference.
- The search functionality for conferences currently available is limited. This can be frustrating as you feel you need to know exactly what you are looking for already.
It would be remiss of me to highlight the issue of finding conferences that have quality content and not recommend those conferences that I do feel are of value.
- ICML — International Conference of Machine Learning
- ECML — European Conference Of Machine Learning and Principles and Practice of Knowledge Discovery in Databases
- CRML — Centre for Responsible Machine Learning
In this article, I have talked about the emotions we feel when going through what can seem like an insurmountable journey to becoming a Machine Learning Expert. I have introduced a resource that can help overcome this lack of expertise that will hopefully only help people in realizing their ambitions. I don’t mean for SlidesLive to be the be-all and end-all of the resources to use, but I feel it is currently underutilized and does not get enough recognition.
Have you previously used SlidesLive? If so, what are your favourite talks/conferences?
- Kruger J, Dunning D.1999, Unskilled and unaware of it: how difficulties in recognizing one’s own incompetence lead to inflated self-assessments. J Pers Soc Psychol. 1999 Dec;77(6):1121–34.