Recent trends in Automated Machine Learning (AutoML) (IN2107, IN4954)

In the last couple of years artificial neural networks (ANN) achieved new state-of-the-art results in a multitude of research areas. Their data-driven training process usually only requires very little in-depth domain knowledge. However, a successful training demands experience and knowledge with respect to ANNs and their design as well as optimization, in other words, an expensive hyperparameter optimization. In a recent trend, research is conducted to facilitate and automate various aspects of this process. In particular, the automatization via Reinforcement Learning is becoming increasingly popular. To this end, we will introduce the most common state-of-the-art RL methods and discuss their application to a selection of modern AutoML papers:

• Learn a gradient descent optimizer
• Neural architecture search (NAS)
• Searching for activation functions
• Learning data augmentation and batching
• …

General information

Date: Wednesdays (11:00-12:00)

Location: Virtual event. Zoom link and password will be shared via email.

Lecturers: Prof. Dr. Laura Leal-Taixé and Tim Meinhardt.

ECTS: 5

SWS: 2

Prerequisites

• Introduction to Deep Learning (I2DL) (IN2346)
• Previous knowledge in Reinforcement Learning preferred.

Course matching

We do not hold a pre-course meeting. Students are supposed to register to the Matching-System between February 11th and 16th. See the Matching-System FAQ for more details.

The matching will take the previous I2DL or DL4CV grades into account.

After the final matching is announced on February 25th, we will send an email with further information and the list of papers. The matched students have time until the first meeting (April 14th) to familiarize themselves with the papers.

Papers

1. Proximal Policy Optimization Algorithms. Schulman et al. (2017).
2. Learning What Data to Learn. Fan et al. (2017).
3. Neural Architecture Search with Reinforcement Learning. Zoph et al. (2016).
4. Learning Transferable Architectures for Scalable Image Recognition. Zoph et al. (2017).
5. AutoAugment: Learning Augmentation Policies from Data. Cubuk et al. (2018).
6. N2N Learning: Network to Network Compression via Policy Gradient Reinforcement Learning. Ashok et al. (2017).
7. DARTS: Differentiable Architecture Search. Liu et al. (2018).
8. Learning to learn by gradient descent by gradient descent. Andrychowicz et al. (2016).
9. Learning step size controllers for robust neural network training. Daniel et al. (2016).
10. Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks. Finn et al. (2017).
11. Meta Dropout: Learning to Perturb Latent Features for Generalization. Lee et al. (2019).

Please feel free to suggest your preferred AutoML paper.

Schedule

• 14.04 (11.00 - 12.00) - Introduction and paper matching
• 19.05 (11.00 - 12.00) - Paper 1
• 26.05 (11.00 - 12.00) - Paper 2 + Paper 3
• 02.06 (11.00 - 12.00) - Paper 4 + Paper 5
• 09.06 (11.00 - 12.00) - Paper 6 + Paper 7
• 16.06 (11.00 - 12.00) - Paper 8 + Paper 9
• 07.07 (11.00 - 12.00) - Paper 10 + Paper 11