Deep learning bioinformatics at Mærsk Tårnet

The teaching fell on 30 May and 1 June 2018. The host was Gopal Karemore at the Novo Nordisk Foundation Center for Protein Research, and the University of Copenhagen paperwork gave the purpose in five words: Guest Lecture Deep Learning Bioinformatics.

Teaching took place at Mærsk Tårnet, the tower attached to the Panum complex on Blegdamsvej. At 09:15 on 1 June the visit was in the lift lobby, in front of the wayfinding wall that lists the tenants floor by floor. Auditoriums and teaching rooms sit on the first floor, the Novo Nordisk Foundation Center for Basic Metabolic Research holds six, seven and eight, and a viewpoint sits on the fifteenth.

The second day's slides run from a single artificial neuron to the perceptron, then activation functions, cost functions, gradient descent and backpropagation, coded by hand in Python before any use of TensorFlow. Two MNIST notebooks follow, one a plain network and one convolutional. A third builds LSTM and GRU layers in Keras over 13,642 blood glucose readings and asks the model for the next value. Alongside them sit the Pima Indians diabetes table, 768 rows and eight columns, and a paper on deep learning for protein-ligand binding affinity prediction.

Away from the teaching there was a high floor with a view over the rooftops, and the town on the last afternoon. Nothing from the first day was kept. The date for it rests on the folio from the Avenue Hotel on Åboulevard: arrival 30 May, departure 2 June, room 315, the guest name misspelt as Tarun Singh.

The expenses were filed on 19 July 2018, seven weeks after the event, on the University of Copenhagen form for foreign guests: taxis, meals, a train to the airport and extra hotel charges, ten receipts totalling 1,445.69 kroner.