📝 Editorial 

Very few projects have had most direct influence in natural language understanding (NLU) and automatic speech recognition (ASR) research than Amazon’s Alexa. Undisputedly considered one of the top two digital assistants in the world, Alexa is used in highly heterogenous devices from mobile phones to cars. The ubiquitous adoption gives Alexa a unique footprint to apply NLU and ASR methods across different languages, environments and interaction modes. Not surprisingly, Alexa has become one of the top environment for evaluating and deploying new ASR-NLU ideas.  

The breadth and depth of NLU-ASR research applied in Alexa challenges is mindboggling. From interactions in low-resource languages, multi-task models to insanely advance methods to improve the quality of speech, Alexa applies hundreds of NLU-ASR models at scale. Just last week, Amazon Research published three new papers about NLU-ASR research applied in Alexa. One of the papers focuses on recognizing intents in unexplored domains which is super relevant to establish broader conversations. Another paper covered the difficult area of dynamic personalization that could be used to tailor a conversation where Alexa learns more info from a client. Finally, the third paper proposes a graph neural networks method to improve the matching of recipes to customer requests.  

This is just one week in NLU-ASR research applied to Alexa. Without a doubt, Alexa has become an important factor and one of the most fertile environments for accelerating NLU-ASR research in the current market.  

🔺🔻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#201: we explain Graph Convolutional Neural Networks; overview the original GCN Paper; explore PyTorch Geometric, one of the most complete GNN frameworks available today. 

Edge#202: we discuss model development vs model deployment, and +how to use Baseten to ship ML-powered apps.

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

🔎 ML Research

New Alexa AI Papers 

For the ICASSP conference, Amazon Research published a number of papers detailing different advanced NLU techniques applied in Alexa devices →read more on Amazon Research blog

Finding Critiques in Summaries 

OpenAI published a paper describing a technique that find flaws in text summaries →read more on Open AI blog

Federated Learning for Data Protection  

Meta AI published a paper detailing a federated learning architecture used to protect data in mobile devices →read more on Meta AI engineering team blog

Direct Speech-to-Speech Translation 

Meta AI Research (FAIR) published a paper proposing a technique for direct speech-to-speech translation without using any text intermediaries →read more on Meta AI Research blog

📌 Event: June 29th – Arize:Observe Unstructured

Arize:Observe Unstructured is open for registration! Learn about emerging embedding techniques from OpenAI and Hugging Face and get your hands on new cutting-edge embedding drift monitoring tools in a fun workshop.

REGISTER HERE

🤖 Cool AI Tech Releases

OmniXAI  

Salesforce Research open-sourced Omni eXplainable AI (OmniXAI), a library to improve the explainability of ML models →read more on Salesforce Research blog

TaiChi 

Salesforce Research open sourced TaiChi, a library for few-shot NLP models →read more on Salesforce Research blog  

🛠 Real World ML 

Offline Inferences at Uber 

Uber discusses the architecture to enable large scale batch inference workflows →read more on Uber engineering blog

NLP in AWS Book 

AWS engineers published a new book about best practices for implementing NLP solutions using the AWS ML stack →read more on Amazon Research blog

GNNs at Airbnb 

Airbnb discusses the architecture and techniques used to apply graph neural networks (GNNs) to different scenarios in the hospitality giant →read more on Airnbnb engineering blog

💸 Money in AI

ML&AI&Data

• ML development platform Lightning AI (ex-Grid.ai) raised $40 million in a Series B round led by Coatue. Hiring in New York/US.

• Data analytics solution Workstream raised $7 million in seed funding led by Lerer Hippeau. Hiring in Atlanta, New York, remote/US.

• MLOps startup Spell.ml is acquired by Reddit for an undisclosed amount. Hiring in New York/US.

AI-powered

• Market intelligence company AlphaSense, raised $225 million Series D financing round led by the Growth Equity business within Goldman Sachs and Viking Global Investors. Hiring globally.

• Conversation intelligence platform Invoca raised $83 million in Series F financing round led by Silver Lake Waterman. Hiring remote. Internship positions are available.

• Loan management platform Able raised $20 million in a Series A funding round led by Canapi Ventures.

• Drug discovery company META Pharmaceuticals, raised $15 million in Seed and Pre-A funding rounds.