2-Min AI Newsletter #9

The 2-Minute AI Newsletter is all about research related to artificial intelligence (AI), machine learning, and deep neural networks. We'll be sharing some of the latest advances in these exciting fields that will help you stay on top of what's new! As always, we won't waste any time, so get back to your busy day while keeping up with knowledge within this edition free newsletter.

???? ?? Latest AI/ML Research Highlights?

???Researchers At MIT Developed A Machine Learning Model That Can Answer University-Level Mathematics Problems In A Few Seconds At A Human Level

• An MIT multidisciplinary research team has created a neural network model that can quickly and accurately answer college-level arithmetic problems.
• The model also automatically explains solutions and rapidly generates new problems in university math subjects
• This work could be used to streamline content generation for courses, which could be especially useful in large residential courses and massive open online courses (MOOCs) that have thousands of students.

???Baidu and BioMap AI Research Open-Sources HelixFold-Single: An End-To-End MSA-Free Protein Structure Prediction Pipeline

• They propose HelixFold-Single, an end-to-end MSA-free protein structure prediction pipeline
• A large-scale PLM serves as the model’s foundation, and the second crucial element consists of the folding-related fundamentals from AlphaFold2.
• The team states that HelixFold-Single outperforms MSA-based techniques in prediction efficiency and could be used for protein-related tasks requiring a large number of predictions.

??Amazon’s 20B-Parameter Alexa Model Sets New Marks In Few-Shot Learning Along With Low Carbon Footprint During Training (One-Fifth of GPT-3’s)

• Amazon has introduced Alexa Teacher Models (AlexaTM), which are massive transformer-based multilingual language models.?
• The team has advanced their study with a 20-billion-parameter generative model called AlexaTM 20B in a companion publication that will be released soon.
• The studies described in the study, which only employ publicly available data, demonstrate that AlexaTM 20B can learn new tasks from a small number of instances and transfer what it learns across languages (few-shot learning).

???Google AI Researchers Propose N-Grammer for Augmenting the Transformer Architecture with Latent n-grams

• This paper suggests a simple alteration to the Transformer architecture, called the N-Grammer.
• During training and inference, the N-Grammer layer uses sparse operations only.
• This work discovers that while being substantially faster at inference, a Transformer framework integrated with the latent N-Grammer layer can also attain the worth of a larger Transformer.

???Startup Featured For This Issue

???Meet ‘Bobidi,’ An AI Startup That Is Helping Businesses Validate Their Machine Learning Models Before Deployment

• Businesses frequently lack the needed time and resources to validate their ML systems and ensure they are bug-free.?With its AI model test platform,?Bobidi, a California-based firm, is attempting to address this issue by enabling businesses to validate models before deployment securely.
• In order to test models and identify biases, it uses its global data science community, making the entire process ten times more effective.
• Founded in 2020, the early-stage startup has raised $5.8 Million in Seed funding.