ThinkML

ML4Creativity - Workshop @ KDD-2017, Halifax, Canada

Located in a suite facing the scenic ocean, this workshop started very early in the morning at 8am. We had four interesting and highly diverse (Gaming, Flavours, Fragrance, Culture … ) keynote talks,

• Mark Riedl, Learning to Make Stuff

• Eric Bruno, The World of Smell

• Flavio Calmon, Challenges in Industry Level growth of computational creativity"

• Nick Montfort, C-Creativity

Along with the keynotes, we had four interesting (not so diverse!) oral presentations,

• Event Representations for Automated Story Generation with Deep Neural Nets

• Story Generation from Sequence of Independent Short Descriptions

• Evaluating Story Generation Systems Using Automated Linguistic Analyses

• mAnI: Movie Amalgamation using Neural Imitation; Visualizing the Movie while Reading a Book

Key Observations as an Audience:

Some of the key observations, learnings, and take-aways, that are learnt from this workshop are:

1. Definition of Computational Creativity is still open!
2. All the four oral papers were related to story generation!
3. Context is the key for getting creative!
4. Automated evaluation is hard. Human evaluation is still the practice!
5. Current applications of machine learning are design aides and not designers
6. Creativity at industry/ enterprise scale is highly challenging
7. Culture has huge implications in building creative systems!
8. Goal formulation and curiosity are essential for creative AI systems
9. Intentionality is critical for creative work

Key Learnings as an Organizer:

1. Number of paper submissions < < Number of attendees! Though there is an explicit interest in the community, far few papers flow by in the broader field of computational creativity, due to the inherent challenging nature of the problem
2. Most of the papers were around story generation, due to the availability of public datasets. Hence, public benchmark data is imperative to perform research in this domain.
3. There is a huge gap between the artistic interpretation and the computational interpretation of creativity! This is the first problem that needs to be addressed in this community.
4. Evaluation is highly subjective and biased. More often, evaluation criteria are made flexible to show enhanced results.
5. Most logical path ahead would be to invest time in creative a huge benchmark dataset and organize a competation/ cup in the next conference!

Having said these, it was really an exciting couple of hours. Learnt a lot about different applications and implications of creative systems, that would have taken weeks for me to catch up browsing through literature. Looking forward to meeting all those interesting folks again, soon!