Profile of TAs

To help you decide with whom you should book a help session, here is a list of the specific expertise our TAs have:

Sofia Broomé

Brief Bio: Third year PhD student at RPL working on detecting pain expressions in horses from video data. 

Deep learning software packages:

• Mainly Tensorflow (including tfrecords)
• But also some Pytorch
• Keras

Deep learning methodologies, architectures and application fields:

• Anything that can model time dependencies in vision: convolutional LSTMs, 3D convolutions, CNNs with RNNs stacked on them, etc.
• Interpretability for deep video models
• Recently looking into multi-view literature
• Some experience with segmentation
• Some experience with optical flow
• Some experience with representation learning
• Interested in deep learning theory and explainability in general

Chris Sprague

Brief Bio: PhD student at RPL and SMaRC Links to an external site. working on learning optimal information gathering behaviours in autonomous underwater vehicles.

Deep learning software packages:

• PyTorch
• GPyTorch

Deep learning methodologies, architectures and application fields:

• Generally interested in reinforcement learning and optimal control.
• Applying to learning optimal environmental monitoring behaviours.
• Recently interested in physics-informed neural networks, Bayesian optimisation, and meta-learning.
• Typically using multi-layer perceptrons for control, but increasingly using PointNet for perception.

Matteo Gamba

Breef bio: PhD student at RPL, working on studying the properties of the functions learned by deep networks for image classification

Deep learning software packages:

• Pytorch!

Deep learning methodologies, architectures and application fields:

• I am working on the general problem of identifying the inductive bias of ConvNets by studying hyperplane  arrangements and activation regions of trained networks
• Anything concerning complexity measures (in the statistical learning sense) and their relationship to implicit regularization/capacity control at play during training
• Interested in deep learning theory and interpretability in general

Lennart Alexander Van der Goten

Brief Bio: First year PhD student at CST (SciLifeLab) working on domain adaptation & anonymization on biomedical data (CT, MRI etc.) 

Deep learning software packages:

• PyTorch
• TensorFlow

Deep learning methodologies, architectures and application fields:

• Generative Adversarial Networks
• Variational Autoencoders
• Large-Scale & Distributed Deep Learning
• DL on volumetric data

Erik Englesson

Brief Bio: PhD student at RPL, investigating efficient uncertainty estimation for image classification. 

Deep learning software packages:

• Pytorch
  

Deep learning methodologies, architectures and application fields :

• Uncertainty Estimation: calibration, out-of-distribution detection, adversarial examples
• Soft Labels: knowledge distillation, mixup, label smoothing, entropy regularization etc
• Meta: Recently, I have also been looking into meta-learning approaches

Federico Baldassarre Links to an external site.

• PyTorch, by far my favorite
• Keras/TensorFlow, I can probably answer high-level questions

• I like everything concerning computer vision and I do my best to be up to date on all CV topics, like dense detections, action recognition, pose estimation, image generation, relationship detection etc. 
• My overall PhD project is about explainability and interpreting the predictions of deep neural networks
• I like graph networks, I keep up with the literature on this topic and I've applied them to molecules and proteins, with also the idea of using relational models to improve interpretability.

Ruibo Tu

Brief Bio: PhD student at RPL, working on causal discovery and missing data problem. 

I was a student of this course during my master study.

If you have questions about

• understanding the content of lectures,
• details about implementing assignments, 
• proposal ideas for the project,

please contact me via my email: firstname at kth dot se. 

For the help session, it would be more efficient  for our meeting if you could let me know your question before the meeting. It would be better for me to prepare for your question if you could book an appointment with me at least one day before the help session.

Deep learning software packages:

• Tensorflow and Keras, the same level as if you finished the course.

Deep learning methodologies, architectures and application fields :

• Interested in the identification problem and the disentanglement representation learning  of VAE.
• Interested in AutoML.

Sebastian Bujwid

• Tensorflow (1.x & 2.x)
• PyTorch

• Multimodal data: text data, CNN, attention models
• Generative models (GAN, VAE, image-to-image translation)
• Computer vision applications: semantic segmentation, depth estimation, etc.

Yiping Xie

Brief Bio: PhD student in RPL and WASP, working on learning the representations of underwater perception for navigation. 

Deep learning software packages:

• PyTorch

Deep learning methodologies, architectures and application fields:

• Generally discriminative models and generative models (VAEs, GANs)
• Computer Visions applications: depth estimation and etc.