Research

• Investigating Representation Collapse in Reinforcement Learning Agents from Vision
  • plan/structure?
  • what RL algorithms?
    • visual data
    • Gehart Neumann, Marc Toussaint, Joustus Piater (Innsburg)
    • Define a research question
    • Focus on some domain
    •  
• Unnormalized Contrastive learning
  • All CL models use l2 normalization of the representation

    • Stability: Normalizing the representations ensures that they all have the same magnitude. This can make the learning process more stable, as it prevents the model from assigning arbitrarily large or small magnitudes to the representations.

    • Focus on direction: By constraining the representations to have a fixed magnitude, the learning process focuses on the direction of the vectors in the embedding space. This is often what we care about in tasks like contrastive learning, where the goal is to make the representations of similar inputs point in similar directions.

    • Computational convenience: As mentioned earlier, many computations, such as the dot product between two vectors, are easier to perform and interpret in normalized spaces.

    • Interpretability: Normalized representations are often more interpretable, as the angle between two vectors can be directly interpreted as a measure of similarity or dissimilarity.

  • BUT, this come to the expense of
    • Decreased Capacity: With normalization, the model’s capacity to represent data is reduced since it can only rely on the direction of vectors in the embedding space. This limitation may result in the model being less able to capture complex patterns in the data.

PhD Registration

• registered

M.Sc. Students/Interns

• Iye Szin presenting next week her work until now.

ML Course

• Publish Video Tutorial on pytorch

Miscellaneous

• Summer School in Cambridge
  • Poster?

Research

• reviews for ECAI (2/6) (Vedant is working on one of them)
  
• Research leads:
  1. Dimensionality Collapse of Visual Representations in Reinforcement Learning
  2. Improve SwAV architecture by creating better latent space clusters with the use of Sparse Autoencoders

PhD Registration

• waiting for admission office response

M.Sc. Students/Interns

• Iye Szin has a working prototype

ML Course

• Tutorial on pytorch
• pending grading for assignments 5 and 6

Miscellaneous

• Summer School
  • Registration done
  • Air tickets booked
  • accommodation booked
• English course got postponed

Research

• submitted CR-VAE paper to ECAI
• Research leads:
  1. Dimensionality Collapse of Visual Representations in Reinforcement Learning
  2. Improve SwAV architecture by creating better latent space clusters with the use of Sparse Autoencoders

PhD Registration

• Signed Application
• Will hand it over to the Admissions office

M.Sc. Students/Interns

• Possible PhD position for Iye Szin
• Early June first draft presentation

ML Course

• Assignment 5

Miscellaneous

• Kleinwassertal

Research

• reviewing paper for IROS
• working on CR-VAE paper
• experiment for the SL competition

PhD Registration

• waiting for Toussaint’s response
• maybe contact other professors?
  • rudolf

M.Sc. Students/Interns

• Iye Szin
  • SL competition; deadline May 1

ML Course

• grades for assignment 2 out

Miscellaneous

• Summer Schools
  • ProbAI accepted (registration until 26/04)
  • ETH & RLSS waiting list
• May-June Leaves
  • 19 May
  • 30 May – 6 June
• Move May 1 to May 10 vacation

Research

• Plan to participate in the air hockey challenge
• Literature review for the right model

PhD Registration

• todo: prepare Email

M.Sc. Students/Interns

• Iye Szin work plan
• Internship will lead to her thesis

ML Assistantship

• Syllabus
• Prepare exercises 

ML Course

• Moodle to upload files (discussed)
• Link to latex for the report (done)

Miscellaneous

• No time to attend the research seminar, ML course takes too much of my time. (discussed)
• 2 days work from home 31.05 & 01.06
• Vacation 02.06 – 11.06
• Medium GPUs for WS in the lab (RTX 3060 or 3070)