1 of 7

Embedded Learning and Sensing Systems GroupASAI networking event @ICML’24

Olga Saukh, saukh@tugraz.at

Embedded Learning and Sensing Systems Group, teams.iti.tugraz.at/elss

Institute of Technical Informatics

Graz University of Technology

AI & Machine Learning

Complexity Science Hub

Embedded Learning & Sensing Systems

2 of 7

Group: Embedded Learning and Sensing Systems (ELSS)

Olga Saukh�(group leader)

Francesco Corti�(PhD student)

Nam Cao�(PhD student)

Katarina Petrovic�(PhD student)

Dong Wang�(PhD student)

Franz Papst�(PhD student)

3 of 7

4 of 7

Vision: Towards Trusted Edge Intelligence

Less is More. Small can be Mighty.

Adaptation / reconfiguration

  • Model reconfiguration
  • Continual learning
  • On-device learning

Trustworthy AI

  • Adversarial �attacks, patches,�and defenses
  • Interpretability
  • Safe AI

Reality “check”

  • Building prototypes
  • Running experiments
  • Reverse-engineering

Optimization for DL

  • Monoliths vs multitudes
    • Ensembling, editing, merging, mixture of experts
  • Model compression
  • Data compression
  • Foundation models / AI agents

5 of 7

Selected Topic: Linear mode connectivity of independently� trained deep models in the weight space

Permutation ConjectureTaking permutations into account, there is likely no barrieron the linear interpolation between SGD solutions

Permutation invariance

C

A

B

A

B

C

Models are invariant to the permutation of neurons as long as the incoming and outgoing weights are permuted too

6 of 7

Selected Topic: Reconfigurable deep learning for edge intelligence

Due to severe resource constraints edge devices often lack support for on-device learning / backprob

Can a model be reconfigured instead of adapted?

-10°

+10°

D. Wang et al., Subspace-Configurable Networks, CoLLAs 2024 �(Oral presentation)

We train a low-dimensional configuration subspace (𝜷-space) used to construct inference network as a linear combination of base model weights

Subspace-configurable �networks make invariances explicit

7 of 7

Join us to discuss recent advances

… and help to spread the word

public