TheSequence Scope: The Mismatch Between Machine Learning Research and Implementation
Weekly newsletter that discusses impactful ML research papers, cool tech releases, the money in AI, and real-life implementations.
📝 Editorial
One of the things that we try to do with TheSequence newsletter is to make machine learning more approachable to developers, data scientists, as well as technologists. Part of that objective requires differentiating between the advancements in AI research and what is really practical with today’s mainstream technologies. In the market’s current state, AI research is advancing at a much faster pace than the frameworks and platforms data scientists use to build solutions in the real world. As a result, many of the ideas we read in research papers prove impractical to implement with modern technology stacks.
Sophisticated models are an important component of machine learning solutions but it’s far from being a key to success. In the current early stage of machine learning platforms, a lot of the magic of real-world solutions relies on infrastructure aspects such as model compression, serving, training optimization, monitoring, and many others. It is incredibly common to find machine learning models that perform great in lab environments but are simply impossible to operationalize. Robust machine learning infrastructure and processes are by far the most important differentiators between success and failure in real-world implementations. Hopefully, we can help increase awareness of the frameworks and best practices that will help machine learning practitioners bridge the gap between research and implementation. We dive deep into explaining concepts, impactful research papers, and useful frameworks twice a week in TheSequence Edge.
🗓 Next week in TheSequence Edge
July 21, Edge#5: Mobile Deep Learning; MobileNets, one of the first mainstream architectures for mobile deep learning applications; and PyTorch Mobile, one of the coolest mobile deep learning frameworks in the market.
July 23, Edge#6: Federated Learning; the Original Federated Learning Paper; and TensorFlow Federated, which enables the implementation of federated learning models on top of TensorFlow.
To stay up-to-date and receive TheSequence Edge every Tuesday and Thursday, please consider joining our community. Till August 15 you can subscribe with a permanent 20% discount ➡️
Now, let’s review the most important developments in AI research and technology this week.
🔎 ML Research
Transfer Learning Metric
Amazon Science published a paper proposing a metric to measure the transferability of models ->read more in this blog post from Amazon Science
Deep Learning Computational Limits
MIT researchers published a paper quantifying the need for more efficient deep learning methods ->read more in the original research paper
Mapping Natural Language Instructions to Actions in Mobile Phones
Discovering new features and actions in mobile apps is a time-intensive process. Researchers from Google published a paper proposing a method that maps language instructions to executable actions end-to-end in mobile apps ->read more in the Google AI blog
🤖 Cool AI Tech Releases
Captum & Fiddler Integration
Facebook made its PyTorch interpretability model Captum interoperable with AI explainability platform Fiddler ->read more in this blog post from the PyTorch team
Microsoft spins out its chatbot Xiaoice
Five years is old enough to be separated from a mother. At least from a mother company, Microsoft decided. The corporation set its chatbot Xiaoice (that many Chinese call their ‘virtual girlfriend’) as an independent entity, in order to accelerate Xiaoice’s “localized innovation” in China, Japan, and Indonesia, and to deal with accelerating Chinese censorship ->read more in the article from TechCrunch
Fabricius by Google
Google created Fabricius, a tool that uses the power of AI to help decode ancient languages. You can learn about Egyptian sign language, send coded messages in hieroglyphs, or assist researchers with translation ->read more on Google Arts & Culture
💬 Useful Tweet
💸 Money in AI
An ML startup Abacus.AI (previously known as RealityEngines.AI) raised $13 million in Series A round. The startup takes the heavy lifting of learning how to train models and helps companies easily implement modern deep learning systems into their business processes.
Robotic process automation may sound extremely boring. But it’s a huge business. UiPath, a startup that automates monotonous, repetitive chores traditionally performed by human workers, has just closed a $225 million funding round, bringing its total raised to over $1.2 billion.
Paige, a health care startup that helps diagnose cancer using computer vision trained on clinical imaging data, has just raised an additional $20 million. They are actively hiring in the US and internationally.
Something sweet in the end. Robotic beehive startup Beewise raised $10 million to provide beekeepers with a management solution combining computer vision, robotic arms, sensors, and software, which takes care of bees in real-time while collecting data for further analysis.
If you find our newsletter useful, please consider supporting our efforts. Subscribe or make it as a gift for those who can benefit from it. It’s a permanent 20% discount until August 15.
TheSequence is a summary of groundbreaking ML research papers, engaging explanations of ML concepts, exploration of new ML frameworks, and platforms. It also keeps you up to date with the news, trends, and technology developments in the AI field.
5 minutes of your time, 3 times a week – you will steadily become knowledgeable about everything happening in the AI space.
