Deep Learning: Self-Supervised Learning for Zero-Shot Tracking

Dear Deep Learners,

We kindly invite you to our next Deep Learning meetup on June 19th, hosted by Bosch.

Our topics will be Self-Supervised Learning for Zero-Shot Tracking.

***

Agenda

18:30

• Introduction by the meetup organizers
• Welcome by the host

18:45

About Bosch

Cancelled

• Talk : Unravelling complexity in time-series data with BunDLe-Net
  Akshey Kumar, University of Vienna

• this talk unfortunately got cancelled

19:00

• Self-Supervised Learning for Zero-Shot Tracking
  Charlie Fieseler, University of Vienna
• Networking & Discussions

20:00

• Announcements: Events & Job Openings

• Networking Break & Discussions

21:30 Wrap up & End

***

Talk Details:

Talk 1: Unravelling complexity in time-series data with BunDLe-Net
Akshey Kumar, Research Group Neuroinformatics, University of Vienna.

Understanding complex dynamical systems, such as the brain, is challenging due to the high dimensionality of the data and complex interactions between components. Even a complete characterisation of cause-effect relationships between variables may not improve our understanding due to the sheer complexity. Conventional dimensionality reduction techniques such as PCA and t-SNE are unsupervised and generally aim to optimise reconstruction quality.

This is limiting when we are interested in understanding the dynamics of a specific target variable. Here, we introduce BunDLe-Net, a manifold-learning algorithm that effectively preserves relevant information while abstracting away details that are irrelevant to the target variable. We apply it to neuronal data from the roundworm C. elegans. BunDLe-Net reveals clear orbit-like trajectories which are recurrent and structured. I like to think of them as 'thought trajectories' since they are derived from neuronal data.

From these trajectories, one can directly read-off information about decision-making, uncertainty, future dynamics and behavioural patterns. It is a powerful visualisation tool for high-dimensional time-series data in the context of a target variable, outperforming conventional and state-of-the-art methods.

Talk 2: Self-Supervised Learning for Zero-Shot Tracking
Charlie Fieseler
Department of Neuroscience and Developmental Biology, University of Vienna, Vienna Biocenter (VBC)

Object tracking is a difficult problem with applications in many domains, from self-driving cars and fundamental biology. The classic supervised learning approach requires large amounts of training data, which is especially difficult for small academic teams with novel datasets. Recently there have been many attempts to reduce this requirement, and one of the most promising is self-supervised pretraining with supervised fine-tuning. This talk will have 3 parts: an overview of self-supervised learning with images, a discussion of practical difficulties reproducing published work with reasonable resources, and a concrete application to cell tracking using a novel extension of the Barlow Twins loss function.

Hot Topics & Latest News:
Do you have some interesting breaking news about Deep Learning? Did you read an interesting paper that you want to share? Did you create an exciting application or achieve some break-through? It would be great to share this in our meetup's Hot Topics section! Please get in touch through contact@vdlm.at

We are looking forward to seeing you at our next meetup!