We ended last year with participating at Codiax, Cluj, a conference focused on Machine Learning and Artificial intelligence. Three of our colleagues participated, and now they are here, answering some questions about their experience.
What is your first impression of Codiax?
Daria: Codiax was the first serious tech conference I attended, but certainly won’t be the last one! I was curious about the event, and it exceeded my expectations since the first speaker started his presentation. I really enjoyed the event especially because of the amazing new people I connected with and all the knowledge I’ve gained.
Darius: This was the first time I attended any Machine Learning and AI conference. This one is a spin-off from Techsylvania which is another Cluj based tech conference. The first impressions kept getting better and better. From just reading about the event, what impressed me was the panel of speakers. It was so diverse and reached all the corners of this domain. The event progressed naturally and what started from general knowledge about ML and AI and the solutions available gradually got more specialized with talks about image preprocessing for efficiency in Neural network models, image generation techniques with ML, synthetic data and even the Metaverse. The speakers were helpful and stayed in proximity in case anybody had any questions. All in all, the conference was very well rounded right from the beginning, did not give you any time or reason to get bored or distracted and had a unique advantage in what it offered at least one subject of interest for any participant.
Georgiana: Yes, this was my first time attending the Codiax conference and I can say that it was a great opportunity to gain knowledge in AI and Machine Learning domain. I was curious and then impressed about how the topics were tackled, I found interesting presentations, useful information and many valuable lessons from the speakers' experiences along with a good ability to deliver captivating talks.
How would you describe the conference in 5 words or less?
Daria: Extensive knowledge and innovative thinking.
Darius: Excellent topics, world-class speakers.
Georgiana: Innovation in technology.
Which were the most interesting topics that you participated at and why?
Daria: One of the most interesting talks was “Artificial Intelligence Based Solution for Video Incident Detection”, presented by Nadiya Shvai from Cyclope.ai. Their solution is implemented in the traffic tunnels of France, tunnels known for their long distance. Using only video-cameras, they can detect if one of the following dangerous situations is happening: a car has an unexpected low speed, a car is on the opposite direction, pedestrians or animals are in the tunnel, a car is stopped or if there might be a fire inside the tunnel. The proposed system leverages different Deep Neural Networks (DNN) based methods to perform object detection, semantic segmentation, and classification tasks.
Darius: The first topic that really captured my interest was that of Julien Simon (Hugging Face) – Hyperproductive Machine Learning with Transformers and Hugging Face. Julien Simon was an Evangelist for Hugging Face which is a company whose main focus is to offer tools in order for users to work with AI easier than ever before, aka they make model-using and -creating extremely easy. You can either use their own models through an interface that resides either on AWS or Azure, but you can also deploy with ease your own models with their tools. To demonstrate he created a model to predict rating scores based on rating review and it took him less than 10 minutes to get it up and running. In my personal experience without proper tools, model deployment and integration is time-consuming and can cause lots of problems, so this is a must have.
The second conference was Soroush Seifi (Automotive Industry) – Visual Attention in Partially Observable Environments. He explained the importance of preprocessing an image before sending it to the ML model. This topic was particularly interesting because he explained in detail side by side “non-processed” and “processed images”. The non-processed ones, work a little better than the other ones, taking into account factors like skies, sidewalks, grass which are for detection and tracking or other applications not so relevant. Because of this, the inference time is increased dramatically. The results with the pre-processed ones (which do not take these factors into account) show a slight drop in performance, but a big decrease in inference time, so overall this approach is a lot better.
Georgiana: The most interesting talk was the one of Bruno Kovacic, CTO at Superbet and Founder at Happening. He described the tight connection Superbet builds between data scientists and the traders following the explosive growth of accessible data and the increase in betting complexity and volume. Integrating the AI in their product helped alleviate the burden of repetitive tasks on humans and managed to reduce the fraud risk in sport betting by spotting suspicious activities and detecting new patterns used in exploiting a sports bet.
Which were the learning you came home with? Or the perspective that made an impact on you?
Darius: What I took from this experience was that this is the time to do a deep dive, create and innovate. There is a second wave of interest in Machine learning (ML) and Artificial intelligence (AI) right now because of new developments in the field. This is what I took home, the inspiration of studying more and starting to ease into the field of AI and ML which will come to the front of the software industry.
Daria: Never stop looking for new and innovative solutions.