Recently I had the opportunity to attend the MLCon conference in Berlin, and I found it to be an enlightening and informative experience. Over two days (out of 4 in total), I attended various sessions and workshops covering a wide range of machine learning and AI topics. In this article, I will share key insights and takeaways from each day of the conference.
Before I move into the conference summary, I would like to thank Arla Foods for making this possible. It’s not the first opportunity I have had with this wonderful employer to explore the trends of tech industry on conferences, update my knowledge, get inspired and network with professionals from industry. You can read more about my previous experiences in my historical articles about EARL and InfoShare conferences in 2021. I really regret not having summarized last year NDSML summit but I highly recommend this event as well if you’re deep into ML world!
MLCon is a premier conference that brings together experts, practitioners, and enthusiasts from the machine learning and AI community. With a focus on practical knowledge and real-world applications, MLCon provides a platform for attendees to stay updated with the latest trends, gain valuable insights, and network with industry professionals.
The conference features a diverse range of sessions, workshops, and keynotes delivered by renowned speakers who are experts in their respective fields. MLCon covers a wide spectrum of topics, including machine learning strategy, product development, deep learning, natural language processing, computer vision, and more.
MLCon is designed to cater to both beginners and experienced professionals, offering a variety of sessions suitable for different skill levels. As you can see on the blurry picture above, majority of participants have been data scientists, but it was also a lot of professionals from software engineering industry, machine learning engineers and business as well! Some called themselved even experienced prompt engineers already what made the whole audience laugh a bit!
One of the key highlights of MLCon is the interactive workshops, where participants have the opportunity to collaborate with industry experts and peers in solving real-world challenges. These workshops provide a unique learning experience, encouraging participants to apply their knowledge and skills in a collaborative environment.
Day 1 – Machine Learning Strategy Day – Sessions
The first day of the MLCon conference focused on machine learning strategy and product development. Expert lectures in the morning provided a theoretical foundation for building successful ML-based products. One of the sessions by Christoph Henkelmann from DIVISIO GmbH highlighted the reasons why ML projects fail. Some of the reasons mentioned were the lack of available data for political or practical reasons, internal conflicts within the team, unrealistic expectations, and not listening to experts.
Henkelmann emphasized the importance of data, continuous iteration and experimentation, and putting effort into data and code cleaning. He also debunked some myths about ML projects, such as the need for fancy MLOps tools or frameworks. He encouraged starting small, communicating effectively, and ensuring that the people working on the project are involved in solving the task themselves.
Peter Buteneers delivered a session on the opportunities and pitfalls of machine learning. He shared interesting examples, such as Facebook training their face recognition algorithm using user-added pictures and the challenge of overfitting in ML models. Buteneers also discussed the potential impact of AI on job markets, self-driving cars, and the importance of human creativity.
Amit Bendor’s session on leading AI teams to the center stage of the company provided valuable insights into managing AI projects within organizations. He emphasized the need to educate clients, partners, and management about AI, manage expectations, and create strong relationships with stakeholders. Bendor also highlighted the importance of positioning AI teams in the organizational structure and building personal connections to get invited to important meetings.
Vinay Narayana’s session on machine learning as an engineering discipline focused on the challenges faced in deploying ML models in production. He discussed the lack of standardized processes and tools for testing, productionalizing, and monitoring ML models. Narayana introduced the concept of MLOps (Machine Learning Operations) and emphasized the need for centralized feature stores, data availability SLOs (Service Level Objectives), and proper monitoring of data and models in production.
Day 1 – ML Product Development Interactive Workshop
During the MLCon conference, we had the opportunity to participate in a highly interactive workshop focused on ML product development. The workshop aimed to provide us with practical insights and approaches to successfully develop ML-based products within our organizations.
The morning lectures provided us with a solid foundation for understanding the challenges and best practices in this field. We learned about the unique characteristics of ML product development and the key factors that contribute to success.
In the afternoon, we were divided into groups consisting of 5-6 participants from different companies worldwide. These groups were intentionally mixed to encourage diverse perspectives and foster collaboration. Each group was required to ideate a specific ML challenge to work on throughout the workshop.
During Step 1 of the workshop, participants engaged in brainstorming sessions within their groups to generate a list of potential ML features for their organizations. They were encouraged to think creatively and identify customer pain points. Understanding the customer’s perspective was emphasized as a crucial starting point for ML product development.
In Step 2, we delved deeper into the chosen challenge and solution. The workshop facilitators provided us with a set of important questions to guide us in documenting the overall solution or feature. These questions helped us clarify the problem statement, define the desired outcomes, and identify the necessary steps to implement the solution. Through group discussions and coaching from ML experts, we gained valuable insights into the practical aspects of ML product development.
By the end of the workshop, each group had a concrete idea of the particular challenges they were facing and the proven approaches for successful ML product development within their organizations. The workshop provided us with a collaborative environment to learn from experts and fellow participants. It empowered us to apply the knowledge gained to our own ML projects.
Overall, the ML Product Development Interactive Workshop was an enriching and engaging experience. It allowed us to put theory into practice and gain practical insights into ML product development. The interactive nature of the workshop fostered collaboration and encouraged us to think critically about the challenges we face in our organizations. It was an invaluable opportunity to learn from industry experts and exchange ideas with professionals from different backgrounds and skill levels.
Day 2 – Conference Day
The second day of the MLCon conference featured insightful keynotes and sessions covering various AI and ML topics. Kathleen Gabriels from Maastricht University delivered a keynote on understanding AI beyond the hype. She discussed the changing definition of intelligence and its mapping onto machines. Gabriels also touched on the challenges of bias in new technologies and the ethical considerations surrounding AI.
Moritz Schrauth and Dr. Alexander Ebbes from Xyna.AI conducted a session on the power of pair programming with AI. They demonstrated how custom prompts can be used to generate specific outputs from language models and discussed the potential applications of this approach.
Jorg Schad from ArangoDB presented a session on connecting large language models and knowledge graphs. He explained the concept of knowledge graphs and their role in representing real-world entities and relationships. Schad discussed the use cases of knowledge graphs in various domains, including logistics, recommendation systems, computational chemistry, and entity resolution.
Dr. Alexander Ebbes from Xyna.AI conducted a session on AI beyond the mainstream, where he discussed the importance of embeddings in representing meaning in AI models. He explained the concepts of normalization, vector addition and multiplication, and scalar products. Ebbes also touched on advanced topics such as autoencoders and retrieval augmented generation (RAG).
Christoph Henkelmann from DIVISIO GmbH introduced us to the concept of embedding vectors. He discussed their usage as inputs for AI models and shared tips for improving classification accuracy using embeddings. Henkelmann also introduced the concept of federated AI learning, where machine learning models are trained on decentralized data while preserving privacy and security.
Pankaj Gupta from DRIMCo GmbH presented a session on OpenFLaaS, which stands for Federated Cloud-Edge AI Rooms for Industrial Knowledge Digitalization. He discussed the challenges faced in implementing AI in German companies due to data privacy regulations. Gupta explained the concept of federated AI learning and its applications in various industries such as healthcare, IoT, manufacturing, and FinTech.
Natalia Ziemba-Jankowska concluded the conference with a session on using audio features to train algorithms. She demonstrated various techniques for sound augmentation, such as time masking, frequency masking, and time warping, and shared Jupyter notebooks available on her GitHub repository for further exploration.
Attending the MLCon conference in Berlin was an incredibly enriching experience. It provided me with valuable insights into the world of machine learning and AI. The sessions and workshops I’ve described in this article are just a selection of the many interesting topics covered at the conference. I would highly recommend checking out the MLCon website agenda for 2023 to explore the full range of sessions and workshops that were available. Also, I deeply encourage all of you to consider attending on your own and grabbing Early Bird tickets for 2024 as soon as those are available!
Let me also acknowledge and express my appreciation for my colleague, Ida Rose Nielsen, from Arla Foods, who attended the conference with me. Our collaboration and discussions throughout the event added another layer of learning and made the experience even more enjoyable.
Additionally, I want to take a moment to appreciate the city of Berlin. Not only did we have the opportunity to gain knowledge and insights at the conference, but we also had the chance to explore the city, indulge in delicious food, and even participate in the Christmas Market on Alexanderplatz. Berlin truly has a unique vibe, and I felt like a Berliner while savoring pretzels and immersing myself in the local culture.
In conclusion, MLCon provided an exceptional platform for learning, networking, and staying up-to-date with the latest trends in the field of machine learning and AI. I am grateful for the opportunity to attend this conference and look forward to applying the knowledge and insights gained to my work at Arla Foods.