The wacky world of AI this week
TensorFlow conference – Google hosted the TensorFlow Dev Summit on Friday in Mountain View, California, and announced a range of updates for its code, which remains the most popular software framework for AI and machine learning.
Here’s a list of the main highlights; all these tools are now going to be rolled out to researchers:
- TensorFlow for Swift: Software libraries for Apple app developers is coming in April.
- TensorFlow Lite: This is for smaller gizmos like mobile phones. It’s core interpreter has been reduced to 75 KB, so its speedier to use.
- TensorFlow RT: Nvidia introduced TensorFlow RT, a software framework optimised for inference using GPUs. It’s also now included in TensorFlow 1.7, the newest version.
- TensorFlow on CPUs: If you’re into CPUs, TensorFlow have partnered up with Intel to include Intel MKL-DNN, another software library geared towards inference.
- Nucleus + DeepVariant: Machine learning is increasingly being used to process genomics data. Nucleus is a library that makes it easier to read, write and filter genomics data files. DeepVariant is a tool to help discover differences in genome code.
- Estimator models: A new method that allows running machine learning models across multiple GPUs using a single machine. It makes it easier to scale up systems without needing to faff around with adding too much extra code.
If you’re a TF nerd, there are more details on the blog and the Youtube channel. Also, here is the download link to TensorFlow 1.7.0 for you to play around with over the Easter weekend.
Je suis désolé Dave je ne peux pas faire ça
La AI stratégie de France – France’s President Emmanuel Macron acknowledged the importance of AI and the need to catch up to the the United States and China.
The French government has pledged €1.5bn ($1.85bn) of public funding to develop AI by 2022 as part of its national strategy laid out in a report by the College de France research institute in Paris.
The translated report said AI was “one of the fascinating scientific endeavors of our time” and the importance of developing systems that were explainable if the technology is to be useful.
It also noted that Silicon Valley is still regarded as the “epicenter for the politics and economics of artificial intelligence”, and that its ‘mindset’ must be adopted in France to accelerate innovation.
“California still dominates in word and in thought and encourages the concept of a single way, technological deterministic approach. If the development of artificial intelligence is fully shaped by private stakeholders, based abroad France and Europe will have no other choice than to their vision.”
“France and Europe need to ensure that their voices are heard and must do their utmost to remain independent. But there is a lot of competition: The United States and China are at the forefront of this technology and their investments far exceed those made in Europe. Canada, the United Kingdom and, especially, Israel hold key positions in this emerging ecosystem,” the report said.
Access to data is key to training AI models. The report called for a policy that will provide “an opportunity to speed up the opening of public data”. But warned that must be in line with the rules in the General Data Protection Regulation that will be enforced on 25 May.
“This data policy must be designed with the aim of safeguarding sovereignty: it is vital for France and Europe to maintain a firm stance on data transfer outside the European Union. The AI strategy must also capitalize on the high protection standards enshrined in the incoming European General Data Protection Regulation (GDPR).”
DeepMind Paris Lab In related news, DeepMind launched a new research lab in Paris led by Remi Munos, a senior research scientist at the French Institute for Research in Computer Science and Automation.
It is DeepMind’s third hub, alongside its headquarters in London, United Kingdom and a lab in Montreal, Canada.
The French team is expected to build upon DeepMind’s work in reinforcement learning. “The DeepMind Paris lab will focus on fundamental AI research, building on Remi’s previous scientific contributions.
These include new state-of-the-art methods that enable single AI systems to learn how to perform many different tasks – a core component of intelligence – as well as fundamental algorithmic breakthroughs such as distributional reinforcement learning,” it said in a blog post.
Nvidia GTC The annual GPU Technology Conference led by Nvidia also took place this week. We covered this in more detail here. CEO Jensen Huang admitted there was a shortage of GPUs, and we did a little digging to discover the wait times for some its chips.
It’s about 12-16 weeks for GeForce cards and 2-4 weeks for the Tesla ones. Read more about that here. ®