Doctor AI, Building an AI OS, Project to Try at Home, What We’re Reading, and More

Issue 58

This week we check out two AI’s that can identify heart arrhythmias and select embryos for fertilization with great accuracy that a doctor can, look at what an operating system for AI would look like, and share what we’re reading, and some projects to try at home.

AI as a Doctor

A team of researchers at Stanford University have a machine learning model that can identify heart arrhythmias from an electrocardiogram better than a doctor.

Researchers collected 30,000 30-second clips from patients with different forms of arrhythmia and trained a 34-layer convolutional neural network, which maps a sequence of ECG samples to a sequence of heart rhythms.

Committees of board-certified cardiologists annotate a gold standard test set on which we compare the performance of our model

Compared to six individual cardiologists, the model exceeded the average cardiologist performance in both recall and precision.

But that’s not all… a collaboration between São Paulo State University and London’s Boston Place Clinic were able to select viable embryos for in vitro fertilization.

Using just 24 key characteristics, such as morphology, texture, and the quantity and quality of the cells present, the AI was able to pick viable embryos 76% of the time.

Between 30% to 60% of seemingly viable embryos fail to implant in the uterus.

AI as an Operating System

As machine learning penetrates the enterprise, companies will soon be able to deploy their models into production and at a faster clip.

But, to run, scale, and monitor hundreds of models in a cloud-agnostic way, you need to think about your AI platform as an operating system, which abstracts the hardware layer from the programming language and technology stack.

This functions-as-a-service architecture enables you to take advantage of efficient, auto-scaling, composable, self-optimizing, and cloud-agnostic infrastructure.

Further, you can maintain interoperability between your model training and inference while abstracting the runtime and the cloud infrastructure at the same time. Win-win-win.

What We’re Reading

  • Benchmarking TensorFlow on Cloud CPUs: Cheaper Deep Learning than Cloud GPUs. There aren’t any benchmarks for deep learning libraries with tons and tons of CPUs since there’s no demand, as GPUs are the Occam’s razor solution to deep learning hardware. (Max Woolf)
  • What Happens When Two Artificial Intelligences Try To Prank Each Other? The friend this user had chosen to prank first was an artificial one, which she had lovingly dubbed “Kaylee’s Robot.” It couldn’t help itself, but respond. (Clément Delangue)
  • Why Artificial Intelligence is Different from Previous Technology Waves. The potential for future innovation with Artificial Intelligence is unlike anything we’ve seen among previous major technology waves. (Robbie Allen)
  • When the automatons explode. As automation becomes cheaper and robotics innovation accelerates, how we work and who we work with will change. (MIT Sloan)
  • How AI detectives are cracking open the black box of deep learning. The deep neural networks are too complex for humans to comprehend, leaving scientists with a nagging question: Why, model, why? (Science)

Things to Try at home🛠

Follow @EmergentFuture for more on frontier technology

Lovingly curated for you by Algorithmia