Two days in Helsinki, from TensorFlow basics to transfer learning
On 5 and 6 June 2019 deepkapha.ai ran a two-day deep learning workshop in Helsinki, taking a room from TensorFlow sessions and graphs through to transfer learning in Keras.
deepkapha.ai ran a two-day deep learning workshop in Helsinki on 5 and 6 June 2019. It was the third Finland engagement in twelve months, after the public sessions of June 2018 and the two-day programme that November.
Day one began with fundamentals. Learners opened a TensorFlow basics notebook, built a single neuron, moved to a multi-layer perceptron, then took on a convolutional network exercise. CIFAR-10 batches sit in the same folder. The solutions notebook carries a timestamp from the following morning, so the exercise was still open when day one closed.
Day two moved to Keras. The room ran MNIST twice, once as a plain network and once as a convolutional one, then put the cats-and-dogs problem through both routes: trained from scratch, and trained again on top of a pre-trained network. Two model files, cats_and_dogs_small_1.h5 and cats_and_dogs_small_2.h5, were written into the learner folder on 6 June and are still there. Recurrent networks closed the day, with a basic exercise and MNIST read as a sequence.
The teaching kit travelled. The instructor folder still holds the deck and the five-day structure built for the deep learning bootcamp at the University of Texas at Dallas, run from 7 to 11 January 2019. The learner folder has only Day 1 and Day 2. Helsinki got the same material at twice the pace.
No photography survives from this visit. Learner-Zone/Photos was created on 4 June 2019 and left empty, so the record here is the teaching material alone.
