The Industrial PhD Scheme
Neddi is participating in Norway's Industrial PhD scheme with Jakob as our doctoral candidate. This program enables companies to collaborate with universities on doctoral projects that address real-world challenges while advancing academic research. The scheme creates a unique partnership where doctoral candidates work on problems directly relevant to their employer's activities while developing specialized research expertise.
Jakob's research at Neddi focuses on developing advanced AI models to personalize learning experiences. His work includes creating intelligent representations of educational content and building student profiles that help teachers identify learning needs early, bridging the gap between cutting-edge machine learning research and practical educational applications.
This collaboration between Neddi and the University of Agder reflects our commitment to combining academic rigor with practical innovation, ensuring that our platform continues to evolve with the latest advances in AI research.
Learn more about the Industrial PhD scheme: https://www.forskningsradet.no/en/financing/what/industrial-phd/
A New Approach to Probabilistic Sampling
The paper, "DESS: Dimensional-PDFs for Embedding Space Sampling," introduces a novel technique for how neural networks generate outputs. Rather than producing a single deterministic answer, DESS trains a neural network to define probability distributions, which enable the model to probabilistically generate variations in the outputs and express confidence scores for the model's own predictions.
This approach is particularly promising for models working with large vocabularies, such as large language models (LLMs) or content recommendation systems, where it could allow substantial reductions in model size and provide new ways to express model certainty estimates. In practical terms, this could mean more efficient AI systems that are both smaller and better at communicating when they're uncertain, a valuable combination for educational applications where trust and transparency matter.
The paper was co-authored with Morten Grundetjern and his supervisors, Per-Arne Andersen and Morten Goodwin.
Open Source and Continued Innovation
In line with Neddi's commitment to open science and collaboration, the code for DESS is publicly available at: DESS-Dimensional-PDFs-for-Embedding-Space-Sampling/README.md at main · neddi-as/DESS-Dimensional-PDFs-for-Embedding-Space-Sampling.
This recognition at SGAI 2025 reinforces the value of combining academic research with real-world educational challenges. We look forward to continuing this work and exploring how these advances can further enhance learning experiences for students and teachers across Europe.
Well done, Jakob! We're thrilled to have you on board.
