📝 Editorial 

In the last few years, PyTorch has unquestionably become (together with TensorFlow) one of the two most popular frameworks for implementing deep learning solutions. A simple and very extensible programming model has made PyTorch a favorite of the data science community. From its early days, PyTorch has quickly added capabilities that address some of the fundamental stages of the lifecycle of ML applications. Last week, PyTorch 1.11 was released with three fundamental additions to the popular framework:  

• TorchData:  A library that abstracts data loading primitives for easily building data pipelines.  

• Functorch: A library that draws inspiration from Google JAX and adds composable function transforms to PyTorch. 

• Distributed Training: Static graph optimizations methods for distributed training. 

PyTorch 1.11 is certainly incremental and doesn’t introduce any major changes in the architecture of PyTorch solutions. However, capabilities such as data loading or composable transformations definitely enhance the experience of building ML pipelines. For now, the race between PyTorch and TensorFlow for the minds of the data science community doesn’t seem to slow down. Check out more details about this new version of PyTorch in the technology section below.  

Share

🔺🔻 TheSequence Scope is our Sunday free digest. To receive high-quality educational content about the most relevant concepts, research papers, and developments in the ML world every Tuesday and Thursday, please subscribe to TheSequence Edge 🔺🔻

🗓 Next week in TheSequence Edge:

Edge#173: we explore Conditional GANs; overview how Meta AI used cGANs to generate images from concepts; explain GAN Lab.

Edge#174: a deep dive into SuperAnnotate, an annotation platform for real-world scenarios.

Now, let’s review the most important developments in the AI industry this week

🔎 ML Research

Restoring Ancient Inscriptions 

DeepMind published a paper discussing Ithaca, a deep learning model that helps historians restore missing text ancient inscriptions →read more on DeepMind blog

Optimizations in Massive Neural Networks 

Microsoft Research published a paper detailing uTransfer, a method for tuning hyperparameters in massively large neural networks →read more on Microsoft Research blog

Robust Graph Neural Networks 

Google Research published a paper proposing a method for addressing some of the bias limitations of graph neural networks →read more on Google Research blog

Estimating Generalization Using Unlabeled Data 

Carnegie Mellon University (CMU) published a paper outlining a technique for estimating generalization using only unlabeled data →read more on CMU AI Research blog

🤖 Cool AI Tech Releases

PyTorch 1.11 

PyTorch released a new version that includes significant enhancements in data loading and manipulations →read more on PyTorch blog

LinkedIn PASS 

LinkedIn open-sourced implementation of Performance-Adaptive Sampling Strategy (PASS) to help researchers build more optimal graph neural networks →read more on LinkedIn engineering blog

💎 We recommend

ML teams face a crowded and complex marketplace for ML infrastructure tools. This "ML Observability Checklist" offers a buyer’s guide with product and technical requirements to consider when assessing an ML Observability platform.

DOWNLOAD NOW

🛠 Real World ML

Safety in Large Language Models 

OpenAI published a detailed blog post discussing some of the lessons learned to maintain safety in large language models such as GPT-3 →read more on OpenAI blog

💸 Money in AI

AI-powered

• Self-driving car solutions company Autobrains raised $120 million in a Series C round led by Temasek. Hiring in Tel Aviv/Israel.

• Data collaboration platform Atlan raises $50 million in a Series B round led by Salesforce Ventures, Sequoia, and Insight. Hiring in the US, India, remote.

• Financial planning and analysis platform DataRails raised $50 million in Series B funding round led by Qumra Capital. Hiring in New York/US, Tel Aviv/Israel.

• Cyber threat intelligence startup Cybersixgill raised $35 million in a Series B funding round led by More Provident, Pension Funds, and REV Venture Partners. Hiring in Tel Aviv/Israel.

• People intelligence company Findem raised $30 million in Series B round led by Four Rivers and Quarry Ventures. Hiring in the US, Canada, and India.

• Quantum computing startup Alice&Bob raised $30 million in a Series A round co-led by Elaia, Bpifrance, and Supernova Invest. Hiring in Paris/France.

• Marketing data analytics startup CaliberMind raised $8 million in a Series A round co-led by IAG Capital Partners and Lavrock Ventures. Hiring in Boulder, CO/US.

*This news digest is presented by Superb AI’s team. We thank Superb AI for their support of TheSequence.

About Superb AI

Superb AI is an advanced DataOps platform looking to transform the way Computer Vision teams prepare and iterate on datasets. The Superb AI Suite provides automation products and tools across all steps of the data preparation workflow, including data labeling, auditing,