Auria Kathi - The power of Multi Model Machine Learning Pipelines

Sleeba Paul (~sleeba) | 

Description:

Multi-Model Machine Learning Pipeline Architecture

Considering the current state of art deep learning algorithms, we might not be able to come up with a single algorithm or network which can build an advanced creative application. But the components of the application can be emulated using individual state of art algorithms. This is called a Multi-model Pipeline architecture for Auria.

Basic outline of the talk

  1. Basic Machine Learning Pipelines [3-5mins]
  2. Using multiple models in a Machine Learning Pipeline [5-7mins]
    • Why multiple models?
    • How it is different from the single model pipelines?
    • What are the challenges?
  3. Case study - Auria Kathi the first poet artist [10-15mins]
    • Auria Kathi introduction
    • Engineering Pipeline of Auria Kathi
    • Azure Machine Learning Pipelines
  4. Conclusion and Q&A session [2mins]

Link to slides: https://bit.ly/2LGP370

Auria Kathi - AI Poet Artist

Auria Kathi is an artificial artist and poet completely living online. "Auria Kathi" is an anagram for "AI Haiku Art". Auria generates a short poem, draws an abstract art based on the poem, and then colors the picture depending upon a mood. All these creative tasks are achieved using a multi-model ML pipeline.

Work of Auria is available in both Instagram and Twitter and will be posting daily for next year.

The engineering pipeline of Auria

  1. An LSTM based language model, trained on 3.5 lakhs Haikus scraped from Reddit. The model is used to generate artificial poetry.
  2. A text to image network, called AttnGAN from Microsoft Research, which converts the generated Haiku to an abstract image.
  3. A photorealistic style transfer algorithm which selects a random style image from WikiArt dataset, and transfer color and brush strokes to the generated image. The WikiArt dataset is a collection of 4k+ curated artworks, which are aggregated on the basis of emotions induced on human beings when the artwork is shown to them.

Engineering Pipeline of Auria

Prerequisites:

Who is this talk for?

  1. Machine Learning Engineers & Data Scientists who are familiar with basic Machine Learning experiments.
  2. The adventurers who would like to attempt complex applications using current SOTA models available by building pipelines.

Key takeaways

  1. A new perspective of seeing Machine Learning pipelines and data flow.
  2. A creative real-life case study like Auria Kathi which demonstrates the application of Multi-model ML pipelines.

Content URLs:

Auria on news and publications

  1. Creative Applications Network - https://www.creativeapplications.net/member-submissions/auria-kathi-an-ai-artist-living-in-the-cloud/
  2. Coding Blues - https://codingblues.com/2019/01/11/fabin-sleeba-and-wonderful-auria/
  3. Creative AI Newsletter - https://us15.campaign-archive.com/?u=c7e080421931e2a646364e3ef&id=d1a15e8502
  4. Towards Datascience - https://towardsdatascience.com/auriakathi-596dfb8710d6
  5. Towards Datascience - https://towardsdatascience.com/auria-kathi-powered-by-microsoft-azure-machine-learning-pipelines-385de55de062

Florence Biennale 2019

At the 12th edition of Florence Biennale happens in October 2019, Auria is exhibiting her work under the contemporary digital art section. Being an international platform for Art, the presence of Auria's work produced by AI will be discussed in Florence Biennale with greater importance. Furthermore, how creative machines are going to build our future by inspiring artists to come up with novel ideas is also a crucial part of the discussion.

Collaboration with Microsoft

Auria is a perfect use case of Microsoft envisioned Azure Machine Learning Pipelines, where each step can be conceived as a containerized computation step. Multiple models developed in diverse environments can be incorporated in the reproducible pipelines and it can be easily deployed as an API. Collaborating with Microsoft, Auria's creative pursuit is coming to a wider audience.

Speaker Info:

Sleeba Paul is a Power System graduate and published researcher who loves intelligent machines. He currently works as a Machine Learning Engineer at Perleybrook Labs; an AI startup in India where he works on video analytics.

Sleeba has research interests in computational neuroscience and artificial general intelligence. He is curious about layers of human emotions and the idea of emotion mappings between people. He believes that empathy in its purest form can be achieved if the emotion of a person can be mapped to another efficiently. Sleeba is a firm advocate of knowledge transfer through stories, especially in the fields of Science, Technology, Engineering, and Mathematics.

Speaker Links:

Section: Data Science, Machine Learning and AI
Type: Talks
Target Audience: Advanced
Last Updated: