WHAT'S IMPORTANT
📍 Timnit Gebru an AI Ethics researcher and co-lead of Google’s ethical AI team, was controversially fired from the company after highlighting risks of large language models and diversity issues inside Google. Fellow researchers, Google employees and private citizens have signed a letter of protest and are voicing support for Gebru under the #ISupportTimnit hashtag. — More on this story at Wired and MIT Technology Review
Image: MIT Technology Review
📍 NeurIPS 2020, arguably the most important Machine Learning conference was held virtually for the first time, boosting attendees from 13,000 last year to a record of 22,000. Fitting with the Timnit controversy, the conference opened with a discussion of bias in machine learning. The Best Paper Awards went to three papers among them OpenAI's paper on GPT-3. — ZDNet wrote an overview of what you missed at the virtual fair.
7 THINGS WE FOUND WORTH SHARING
💡 1. Use Cases - A blind man completed a solo 5K run without any guidance by another human or his guide dog. Instead, Thomas Panek of "Guiding Eyes for the Blind" teamed up with Google's Project Guideline who developed a machine learning model to help the runner navigate on his own. The model run completely on a waist-mounted phone, uses the phone's camera to find the path and audio signals to guide the runner (google.com).
📖 2. Papers - TopBots have selected the 10 most important machine learning research papers of the year 2020. They also provide short summaries for the GPT-3, AdaBelief and Visual Transformer papers, among others (topbots.com).
🎓 3. Education - Machine learning veteran Yann LeCun’s Deep Learning course at the NYU Center for Data Science is now fully online & accessible to all (nyudatascience.medium.com).
👩💻 4. Code - Are your Deep Learning models responding slow under load? Build a smarter API to optimize response time following this Guideline by French Data Scientists at Sicara (sicara.ai).
👩💻 5. Code - Toonify yourself! Create a cartoon character of yourself with StyleGAN2. Ready-to-run Code is available in Google Colab (colab.research.google.com).
💭 6. Articles - Similar to packaged food nutrition labels, IBM's AI FactSheets summarize the capabilities and benchmarks of an AI and contain information about the model’s development. After IBM Research introduced FactSheets in an 2018 paper it is now being commercialized into the Watson Studio suite (youtube.com).
💭 7. Articles - Carl Anderson, Data VP at Weight Watchers, recounts how the company successfully built a data science team from scratch in two years (medium.com).
Want to bring AI to your company? 🧠 Lets talk about it
Any questions, suggestions or feedback? Let us know what you wish for in this newsletter at newsletter@am.ai
Thanks for reading,
- The i.am.ai Newsletter Team
