Budapest AI Meetup - May | RapidMiner, Ericsson, ARGUS Cognitive
Details
- Networking & pizza: 17:40 - 18:00
- Presentation(s): 18:00 -
http://study.inf.elte.hu/aimonday/
- Zoltán Tősér (CEO | ARGUS Cognitive) - When reality hits: designing AI applications for real-world usage
https://arguscognitive.com/
Nowadays most novel technologies revolve around AI. Even Google’s
research division was renamed Google AI last year. It seems inevitable
that once these powerful algorithms become a part of the everyday
software engineer’s toolkit, they will quickly supersede traditional
solutions for quite a few problems. At least until reality hits.
When Andrej Karpathy and Elon Musk presented the latest version of
Tesla’s self-driving system a few weeks ago, they revealed that even
though they had a well-performing vehicle classifier, it would fail when
a bicycle was attached to the back of a van – it labeled it as if it was
crossing the road. And this was not even an adversarial example – those
have demonstrated numerous times that the AutoPilot is easily fooled to
navigate into oncoming traffic.
How should we prepare for such situations in our AI applications? Can we
find good design principles? What about testing? One thing is for sure –
this is not something we can sweep under the rug, or else our AI models
will only work under rather limited circumstances.
-
Tamás Éltető (Product Owner | Ericsson) & Patrik Olesen (Principal Developer | Ericsson) - Machine Learning value, strategy and results at PDU Transport
-
Balázs Péter Hámornik (Data analytics product expert, ex-RapidMiner) - Automatic Predictive Modeling: Machine learning for the business
There is an observable shortage of data science competencies in enterprises globally hence there is a trend that more and more people are becoming data aware, performing advanced analytics, and getting started with data science regardless their field.
We see friction between data scientist people and business or subject matter experts that might be caused by the steep learning curve of data science and its toolset.
Data science product vendors (like RapidMiner, Microsoft, Dataiku) is experienced in supporting data scientist becoming experts in the field. Now the market players is looking forward extending this to business experts (business analysts, data analysts, subject matter experts) to help to develop their data science competencies and enable them to become a citizen data scientist in their companies. However, there are new specialized vendors in this field providing automated ML primarily (DataRobot).
To train a well applicable ML model: you need the right problem identified, the solution/model prototyped, only then it can be operationalized and put the model into production. There are different capabilities are required for each stage.
I think business people should be involved in identifying the right problem and prototype whether predictive modeling or ML can deliver a good solution to it. Then the model(s) can be refined and deployed by data scientists. This enables the company to focus the data scientist resources on the right business problems.