Deep Learning – Generative Networks GANs and Autoencoders
- Deep Learning is taking with world by storm and technologies such as TensorFlow are empowering the age of Artificial Intelligence. This workshop will teach you from a gentle introduction into Artificial Neural Networks and then it will take you into a deep dive into developing your first Neural Network with TensorFlow already!
- This training was ranked #2 in the world out of 8000 trainings, according to the Inc. Magazine. Now it’s available for the first time in Helsinki!
- The instructor Tarry Singh is as CEO of deepkapha.ai, a startup focused on advanced AI research for applied AI for enterprises. He has so far trained over 16,000 deep learning engineers himself in classical mode — in corporate enterprises, educational institutions as well as with humanitarian organizations. He is also mentor at Coursera’s Deep Learning Specialization working with Andrew Ng, the world’s leading figure in Artificial Intelligence as well.
- We will also explore the the necessary skill sets required so you can fast track from a novice, whiz past the world of Machine Learning and get a good grip of Deep Learning by understanding how neural networks work.
We will take a look at the use of Generative Models such as GANs and Autoencoders. We will explore their use in your production environments and we will also talk about the many pitfalls and challenges. Topics covered in the two days:
- Understanding Generative Models and essential building blocks of GANs (Generative Adversarial Networks).
- As fresh new curriculum for 2019, We will do extensive deep learning training on dedicated GPUs (Nvidia 1080, 64GB RAM, SSD flash disks) in GANs and Autoencoders for various exciting datasets.
- Use of Deep Learning models such as GANs in semi-supervised domains and exploring image generation both in creative as well as medical domains.
- Transfer Image Style Generation across various domains with GAN models such as CycleGAN and CGAN. Understand hoe to prevent model collapse by stabilizing your neural network using BEGAN.
- Transferring styles from different domains such as Apple to Oranges, Lion to Tiger, your current business object to another.
- Build realistic Images from Text. Manage multiple subproblems while decomposing text to image with the use of StackGAN or DISCOGAN.
- Taking ML/DL models into production by using microservice containerized or serverless computing.
- Autoencoders –The fundamentals of Autoencoders. Lossless, Lossy and Domain Encoding.
- Build your simple Autoencoder as well as the Digits/MNIST Autoencoder.
- Exploring Autoencoder – Latent Spaces, Parameter Space and Blended Latent Space
- Convolutional Autoencoders and Variational Autoencoders – what are they , learning through coding exercises.
- Finally as a special bonus for the class — you get to learn and possibly even play with the breakthrough activation function ARiA and LayGo controller based learning rate which deepkapha.ai researchers wrote. They have already beaten ReLU and Google’s SWiSH function hands-down as well as other LR algorithms!
- Wrap up and an even harder but fun bonusHomework Exercise — for those who dare
Early registrants will get 5 day extra dedicated GPU server provided by deepkapha pre-installed with all latest deep learning libraries for free. You can train your neural network for unlimited hours.
Please, sign up for training well in advance, so that we can ensure you the capacity used in the course by using our learning platform and server.
Koulutustapahtuman päivittäinen alkaminen ja päättyminen: aamiainen 8:30, koulutus 9:00-16:15
Daily start and end of training event: breakfast 8:30, training 9:00-16:15