📝 Editorial 

Recreating the cognitive patterns of human intelligence has long been the elusive goal of artificial intelligence (AI). Despite the massive progress in deep learning, AI systems remain pretty far from resembling human-like intelligence. One of the hallmarks of human intelligence is our ability to develop models of the world using a lot of background knowledge and very minimum interactions with a given environment. We don’t need to learn every single concept to master a new task. We often refer to this as common sense, and it seems to be one of the missing building blocks of the current generation of AI systems. Recent deep learning disciplines such as self-supervised learning (SSL) seem to have a solid foundation for building AI agents that can develop models of the world relatively inexpensively by reusing existing knowledge.  

In the Meta (Facebook) AI ‘Inside the Lab’ event a few days ago, AI legend Yann LeCun outlined an architecture to build AI agents that can create models of the world in a self-supervised way and then use those models to predict, reason and plan. LeCun’s architecture is based on six core modules, each able to compute differentiable objective functions and pass them to the upstream modules. Not surprisingly, SSL is at the center of LeCun’s ideas as a way to build predictive models of the world. However, each of the six core modules encompasses many areas of AI research. Everything indicates that LeCun’s architecture would evolve as the foundation for Meta’s AI capabilities for metaverse environments. 

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#169: we explain CycleGANs; walk through the original CycleGAN paper; overview Mimicry. 

Edge#170: we deep dive into Superb AI data training platform.

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

🔎 ML Research

An Architecture for Autonomous Intelligence 

As part of the Meta AI’s Inside the Lab event last week, Meta AI Chief Scientist Yann LeCun outlined an architecture for autonomous AI systems which can break the boundaries of traditional supervised learning →read more on Meta blog

What’s Next for Deep Learning 

Amazon Research published an insightful blog post about the integration of neural networks and symbolic systems as the foundation for the next phase of deep learning →read more on Amazon Research blog

Competence-Based RL 

Berkeley AI Research (BAIR) published a paper benchmarking different unsupervised RL methods and introducing a new competence-based RL algorithm that seems to outperform other alternatives →read more on BAIR blog

Translating 100 Written and Spoken Languages in Real-Time 

Meta (Facebook) AI Research (FAIR) published a blog post detailing the evolution of their work in multi-language translation systems for text and speech →read more on FAIR blog

🤖 Cool AI Tech Releases

TorchRec 

The PyTorch team open-sourced TorchRec, a library for building recommendation systems →read more on PyTorch blog

ML Ready Pipelines 

ML startup cnvrg.io announced AI Blueprints, an open-source collection of ML pipelines that shorten the implementation cycle of data science teams →read more on cnvrg.io blog   

💎 We recommend

🛠 Real World ML

Salesforce Research published a blog post detailing the architecture for their app recommendation system →read more on Salesforce blog

🗯 Useful Tweet

Yann LeCun @ylecun

Video of a technical talk I gave: "A Path Towards Autonomous AI" Hosted by Baidu on 2022-02-22. Video: lnkd.in/db_eSSyA 1/Yann LeCun: “A Path Towards Autonomous AI”, Baidu 2022-02-22Technical talk by Yann LeCun:“A Path Towards Autonomous AI”Hosted virtually by Baidu on 2022-02-22.TL;DR: - autonomous AI requires predictive world models- w...lnkd.in

February 25th 2022

56 Retweets247 Likes

We turn our educational Edges into threads and recommend helpful books and courses.

Follow us on Twitter

💸 Money in AI

• Former Google CEO Eric Schmidt committed to investing $125 million over five years into AI2050, an initiative aimed to support AI research with a positive impact. The funding is provided through Schmidt Futures.

• GraphQL developer platform Hasura raised $100 million Series C funding led by Greenoaks. Hiring remote.

• ML Observability Platform Aporia (check the overview in Edge#160) raised a $25 million Series A funding round led by Tiger Global. Hiring in Tel Aviv/Israel.

• Compliance operating system platform anecdotes raised $25 million in Series A funding round led by ‍Red Dot Capital Partners. Hiring in Israel.

• Kubernetes development platform Okteto raised a $15 million Series A funding round led by Two Sigma Ventures. Hiring remote in the US/EU.

• Synthetic avatars creating platform Neosapience raised $21.5 million in a Series B round led by BRV Capital Management. Hiring in Seoul/South Korea.

• Real-time data engineering company Decodable raised a $20 million round led by Venrock and Bain Capital Ventures. Hiring remote.

• Sales engagement platform Vymo raised $22 million in a Series C funding round led by Bertelsmann India Investments.

• Property operations software platform Facilio raised $35 million in Series B financing led by Dragoneer Investment Group.

• Automation company for life sciences Automata raised $50 million in a Series B funding round led by Octopus Ventures. Hiring in London/UK.

• No-code platform Skael raised $38 million in a Series A financing round led by RTP Global. Hiring in the US, Mexico, EU - remote.