📝 Editorial 

Creative expression has been a central aspect of the evolution of humankind and one of the marvels of human cognition. Emulating creativity using artificial intelligence (AI) has been a longtime goal of the industry. Nothing reflects creativity like art which is why, in recent years, we have seen a proliferation of text-to-image, image-to-image generative art methods that have showcased the potential of creative expression in AI models. Methods such as OpenAI’s DALL-E 2 or Google’s Imagen have achieved amazing milestones in generating photorealistic artistic images from natural language inputs. Last week, Meta AI published details around a new technique called Make-A-Scene, which highlighted new dimensions of the potential of generative art models.  

Most generative art techniques are highly specialized in one domain, such as text-to-image or image-to-image. Make-A-Scene takes a multimodal approach by generating artistic representations from both: natural language and freeform sketches. The multimodality adds a level of specificity that is impossible to achieve with previous models, as sketches can often reflect aspects such as position, size, and relationships between objects which are hard to express in natural language. Meta tested Make-A-Scene with leading digital artists such as Sofia Crespo, Scott Eaton, Alexander Reben, and Refik Anadol. All of them were incredibly impressed with the level of control in the artistic creation process and the quality of the outputs produced by the model. Make-A-Scene pushes the boundaries of generative AI models in the fascinating quest to emulate artistic creativity.  

🔺🔻TheSequence Scope – our Sunday edition with the industry’s development overview – is free. To receive high-quality content about the most relevant developments in the ML world every Tuesday and Thursday, please subscribe to TheSequence Edge 🔺🔻

🗓 Next week in TheSequence Edge:

Edge#209: we start a new series about ML testing; explore how Uber backtests time-series forecasting models at scale; discuss Deepchecks, an ML testing platform you should know about. 

Edge#210: we deep dive into Hopsworks 3.0, which bridges the Modern Data Stack with the machine learning stack in Python.

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

🔎 ML Research

Make-A-Scene

Meta AI published a research paper discussing Make-A-Scene, a multimodal generative model that can output artistic representations from text descriptions and image sketches →read more on Meta AI blog

Plex

Google Research published a paper introducing Plex, a framework to improve the reliability of deep learning systems →read more on Google Research blog

Improving Multitask Learning 

Amazon Research published a paper presenting a knowledge distillation method to improve multitask learning models →read more on Amazon Research blog

Visual Question Answering 

Google Research published a paper proposing a method for generating question-answer pairs for images →read more on Google Research blog

☝️ We Recommend

For Tide, a UK business banking provider, the ability to detect fraudulent transactions and make credit approval decisions in real-time is key to business success. Sign up for this webinar where Hendrik Brackmann, VP Data at Tide, and Mike Del Balso, Co-founder and CEO of Tecton, discuss the key challenges that Tide encountered with their first attempt at building a feature store to support their real-time machine learning use cases—and how they solved them. There will be a live Q&A after the presentation. 

REGISTER NOW

🤖 Cool AI Tech Releases

MLU Explains  

Amazon Research released MLU Explains, a website containing visual essays about the core machine learning concepts →read more on Amazon Research blog

BetterTransformer 

PyTorch released BetterTransformer, a high-performance API for transformer inference →read more on PyTorch blog  

Deep Search  

IBM open-sourced Deep Search, a toolkit for searching and discovery across large volumes of unstructured data →read more on IBM Research blog

🛠 Real World ML 

Improving YouTube 

DeepMind published some details about their deep learning work at YouTube in order to improve the experience of the video platform →read more on DeepMind blog

💸 Money in AI

• Feature platform Tecton raised $100 million in a Series C funding round led by Kleiner Perkins. Hiring in San Francisco, New York, Austin/US.

• AI startup AI21 Labs raised a $64 million Series B funding round led by Ahren. Hiring in Tel Aviv/Israel.

• Speech-to-text API platform AssemblyAI raised $30 million in a Series B round led by Insight Partners. Hiring remotely.

• Deep learning development platform Deci raised $25 million in a Series B funding round led by global software investor Insight Partners. Hiring in Tel Aviv/Israel.

• Mental health chatbot Wysa raised a $20 million Series B round led by HealthQuad. Hiring globally.

• AI model test platform Bobidi raised a $5.8 million seed round. Hiring in South Korea.