You are invited to a day full of machine learning

After two successful dida conferences in the last two years (see the video below and recordings of the talks 2023 and 2024) we are looking forward to a new edition this year. The conference will feature talks on machine learning (applied and technical), workshops, space for networking and, of course, good food.

You are invited to a day full of machine learning

This year we have a new event location with the frizzforum in Berlin. As the number of participants is limited, we recommend early registration.

We have a limited number of places available. Please register here for free:

Conference Schedule

  • -

    Doors are open

    Arrive and connect with others.


  • -

    Introduction

    Welcome to the dida conference.


  • -

    Andrey Vasnetsov
    (qdrant)

    miniCOIL: Sparse Neural Retrieval Done Right

    In this talk, we present miniCOIL - our attempt to make a sparse neural retrieval model as it should be - combining the benefits of dense and lexical retrieval without propagating their drawbacks. We will share how to design and train a lightweight model that is performant on out-of-domain data and demonstrate its capabilities.


  • -

    Dr. Sebastian Lapuschkin
    (HHI)

    Decoding AI: How Explainability Unlocks Actionable Insights

    Contemporary Large Language Models (LLMs) constitute a powerful but often opaque enabler technology. This talk explores how tools from Explainable AI (XAI) and mechanistic interpretability can help us uncover the roles of learned structures - such as attention heads or knowledge neurons - enabling more reliable and actionable interactions with AI. We demonstrate how local explainability reveals when an attention head actively contributes to inference and where it draws information from, while global explainability and mechanistic tools uncover its broader function. Using these insights, we introduce computationally efficient methods to detect when an LLM "lies" relative to its own knowledge, identify conflicts between context and parametric memory, and annotate each generated token with measurable reliability scores. Our techniques operate with minimal computational overhead, making real-time, citation-worthy, and controllable LLM outputs feasible in cost-sensitive applications. These advancements push LLMs toward greater transparency, reliability, and practical integration in real-world systems.


  • -

    Coffee break


  • -

    Axel Besinger
    (dida)
    Dr. Augusto Stoffel
    (dida)

    From OCR to LLMs: The Journey to Reliable Data Extraction from Complex Retail Documents

    AI-powered data extraction works - until it doesn’t. When handling structured tables in invoices, orders, or financial documents, we expect OCR, LLMs, and Vision AI to extract data reliably. However, complex documents - e.g. nested tables, irregular structures, and edge cases - pose real challenges for document data extraction AI models. With our solution smartextract, we tackled a real-world customer challenge: automating order entry from complex order documents and tables for a German shoe retailer: OCR and text-based LLMs struggled, Vision LLMs were inconsistent. Only extensive customization could solve the appearing problems - including segmentation, few-shot prompting, fine-tuning, and even the possibility of training a custom computer vision model. In this talk, we will show why standard AI models struggle with complex tables and demonstrate in which cases segmentation helps. Further, we will show benchmarks of commercial vs. open-source models and discuss the trade-offs between OCR, LLMs, and computer vision models.


  • -

    Lunch break


  • -

    Dr. Florian Wenzel
    (Mirelo AI)

    Generating Music and Sound with Generative Foundation Models

    In the age of digital content, music and sound are crucial elements that enhance storytelling, engagement, and emotional impact. Yet, finding the perfect soundtrack for a video - one that aligns seamlessly with its mood and pacing - remains a challenge for many creators. Mirelo AI is changing this with cutting-edge generative foundation models for AI-driven music and sound design. In this talk, we will introduce our approach to video-aware music generation. Our technology enables users to upload a video, and our AI dynamically composes original music and synchronized sound effects tailored specifically to its content. This is a game-changer for content creators, advertisement agencies, game developers, and everyday users looking to elevate their projects with custom-generated audio.


  • -

    Fabian Dechent
    (dida)

    Building a Real Time License Plate Matching Service

    This talk covers strategies applied in bringing a real-time license plate matching service into production. Based on the project requirements, we discuss the chosen ML model architecture and training procedure, implemented software architecture components, as well as tools and infrastructure for the deployment. Particular focus will be on the deployed microservices architecture and model serving using NVIDIA Triton.


  • -

    Coffee break


  • -

    Thomas Strottner
    (Edgeless Systems)

    Confidential AI: Leveraging LLMs and Coding Assistants with End-To-End Encrypted User Data

    With confidential computing, data can remain protected or encrypted even during processing. Since the introduction of the Nvidia H100, this technology can also be applied to AI. When used correctly, it enables the creation of AI-as-a-Service (such as ChatGPT) where users do not need to trust the provider regarding data security and privacy. This makes it possible to process sensitive with GenAI services. The presentation will explain the fundamentals of the technology and provide an overview of use cases and existing deployments.


  • -

    Dr. Ma Li
    (dida)

    Detecting Subtle Events in Videos with YOLO-in-Time

    In this talk, we present a general-purpose model architecture that is suitable for detecting and counting subtle events in videos. It combines a convolutional neural network (CNN), a recurrent neural network (RNN) and a YOLO-type head in the time domain. The model can be trained end-to-end with time-resolved labels only, without the need for customization or domain-specific knowledge. We discuss possible use cases and show experimental results both on synthetic datasets and on real world datasets.


  • -

    Poster Session

    The posters will be announced later. If you would like to present a poster, please send us a message to conference@dida.do.


  • -

    Dinner & Networking

    Food and drinks will be available for all guests.

Workshops: Room 1


  • -

    Dat Win Edwin Fung
    (Wenglor)
    Dr. Tassilo Glander
    (Wenglor)

    Challenges in Model Quantization for an Embedded Platform

    We present our experience of using AI models in the context of industrial computer vision. In manufacturing, AI often is used in a setting of restricted computational power, e.g., on an edge computer or a smart camera without a GPU. In this workshop we will present model quantization as a typical approach to address this challenge, including definition, implementation, pitfalls, and practical use cases. During the workshop, we would like to have a hands-on session so that participants can deploy their own models to an industrial smart camera. Also, we would like to get into an exchange about problems and solutions with the participants.


  • -

    Lunch break


  • -

    Dr. William Clemens
    (dida)

    dida reading group

    dida has a weekly internal reading group where our ML scientists discuss recent papers and how they can help with our projects. The papers we will discuss in this session will be announced later.

Workshops: Room 2


  • -

    Dr. Liel Glaser
    (dida)
    Lorenzo Melchior
    (dida)

    The dida code: Unlocking ethical & secure AI - our path to AI act compliance

    Starting a new project is exciting, defining goals, aligning team expectations and getting the first insights into a new problem. To guide us in this, we have found it effective to hold a kick-off workshop. In light of the recent AI Act, and its added focus on AI and data security, we decided to develop an add-on for the workshop, specifically to classify the project and to cover project specific security requirements. To develop a workshop that helps us develop the best and safest system that we can, without boring our colleagues by repeating the same warning about phishing emails at every project start, we used interviews with past project leads to refine the content, and develop  the workshop. In this talk + workshop we will present how we developed this process, and then try it with you using an example.


  • -

    Lunch break


  • -

    Dr. Marty Oelschläger
    (dida)
    Fabian Dechent
    (dida)

    From Development to Production: Deploy ML Models via Nvidia's Triton Inference Server

    You have developed and trained a nice machine learning model, which works seemingly perfect in your development environment. But now, in the real world of production you encounter various challenges from scaling, over monitoring, to robustness. In this workshop we want to present a step by step tutorial how to use and deploy machine learning models provider agnostic via Nvidia's open-sourced Triton Inference Server as a remedy  for the above mentioned challenges.

Speakers

Andrey Vasnetsov

Andrey is the Co-Founder and CTO of qdrant, where he advances high-performant vector similarity search technology. He has a background in computer science.
Dr. Sebastian Lapuschkin

Sebastian is the head of the explainable artificial intelligence group at Fraunhofer Heinrich Hertz Institute (HHI) in Berlin. He has a PhD in computer science and actively publishes in the field.
Dr. Augusto Stoffel

Augusto holds a PhD in mathematics and did research in the field of algebraic topology and its application as a foundation of quantum field theory. At dida he is a technical lead in information extraction and NLP.
Dr. Florian Wenzel

Florian has a PhD in machine learning and worked as a researcher at Google and Amazon before he founded his own startup Mirelo AI.
Dr. Ma Li

Ma Li has a PhD in pure mathematics and works as a Machine Learning Scientist at dida, focusing both on computer vision and NLP.
Dr. Liel Glaser

Liel has a PhD in theoretical physics from the Niels Bohr Institute in Copenhagen and focuses on NLP as a Machine Learning Scientist at dida.
Thomas Strottner

Thomas is the Vice President of Technology Partnerships at Edgeless Systems.
Axel Besinger

Axel has a passion for business development in technology companies and recently dived into LLMs. He leads smartextract, dida's information extraction software.
Dr. Tassilo Glander

Tassilo has a PhD in computer science and is the CTO of Wenglor Deevio, where he works mostly on computer vision.
Dat Win Edwin Fung

Dat works on machine learning and computer vision problems at Wenglor.
Dr. William Clemens

Will holds a PhD in string theory and quantum chromodynamics at the University of Southampton. At dida, he works as a machine learning scientist.
Fabian Dechent

Fabian is a project lead and machine learning scientist at dida, focusing on both computer vision and NLP.
Dr. Marty Oelschläger

Marty holds a PhD in physics (HU Berlin) focussing on fluctuation-induced phenomena. At dida, he works as a project lead.
Lorenzo Melchior

With a background in computer science, Lorenzo is experienced in dev-ops and data security and works as a Machine Learning Engineer at dida.

Location