Multiscale Vision Transformer for Video Recognition

Multiscale Vision Transformer for Video Recognition

Vision transformers are already good at multiple tasks like image recognition, object detection, and semantic segmentation. However, we can also apply them to data with temporal information like videos. One such use case is using Vision Transformers for video classification. To this end, in this article, we will go over the important parts of the Multiscale Vision Transformer (MViT) paper and also carry out inference using the pretraining model. Figure 1. An example output after passing a bowling video through the Multiscale Vision Transformer model. Although there are several models for this, the Multiscale Vision Transformer model stands out for…
Read More
The long path of JavaScript – from ES6 until today.

The long path of JavaScript – from ES6 until today.

According to a Stack Overflow survey, JavaScript was the most popular language among developers in 2023. JavaScript was initially developed for Netscape Navigator - a web browser that was developed in the middle of 1990s - and now is being used in almost every domain of programming - Web Development, Mobile app development, Game development, Machine Learning and many others. But how did a language which was developed in 10 days by Brendan Eich become so popular? In this article, we will go through the life of JavaScript from ES6, which was released in 2015 and was the second major…
Read More
Another OpenAI Researcher Quits, Issuing Cryptic Warning

Another OpenAI Researcher Quits, Issuing Cryptic Warning

"One of the ways tech companies in general can disempower those seeking to hold them accountable..."Another OneAnother OpenAI researcher left the company this week, ominously citing concerns over opaque "decision-making processes" in the AI industry.In a thread posted this week to X-formerly-Twitter, former OpenAI policy researcher Gretchen Kreuger announced her departure, writing that this "was not an easy decision to make." And while the ex-OpenAIer didn't quite go into detail — can't imagine why not! — about the forces that made her make that difficult choice, she did offer a cryptic warning about the lack of oversight within the AI…
Read More

Extending the RoPE

Rotary Position Embedding (RoPE) is an effective position-encoding technique first introduced in Su et al. (2020) [1] and later popularized in open-source models such as GPT-J, GPT-NeoX, PaLM, LLaMA, etc. We covered the mathematics and the implementation details of RoPE in this blog post about 2 years ago. Although the RoPE is limited by its pretrained context size, we will summarize a line of research that manages to extend the context length of the RoPE so that a pretrained language model can be easily adapted to fit the increasingly challenging tasks being given to LLMs. Given a sequence of tokens…
Read More
The Week’s 10 Biggest Funding Rounds: Scale AI’s Massive $1B Deal Leads Slower Week

The Week’s 10 Biggest Funding Rounds: Scale AI’s Massive $1B Deal Leads Slower Week

Want to keep track of the largest startup funding deals in 2024 with our curated list of $100 million-plus venture deals to U.S.-based companies? Check out The Crunchbase Megadeals Board. This is a weekly feature that runs down the week’s top 10 announced funding rounds in the U.S. Check out last week’s biggest funding rounds here. It was a slower week for big rounds compared to many of the past few. Nevertheless, the week saw a $1 billion round go to yet another AI startup, and biotech and blockchain also saw some big deals. 1. Scale AI, $1B, artificial intelligence:…
Read More
Snowflake Looks to AI to Bolster Growth

Snowflake Looks to AI to Bolster Growth

(Michael Vi/Shutterstock) Investors in Snowflake breathed a sigh of relief this week when the cloud data warehouser reported solid revenue growth for its first quarter and raised its guidance for the rest of the year. But questions still remain over its long-term growth, which the company is hoping that artificial intelligence will power. The company’s acquisition this week of assets of TruEra fits that mold. Snowflake on Wednesday reported $829 million in total GAAP revenues for the quarter ended April 30, 2024, representing a 33% increase over the same period last year. It reported 14 cents per share, which was…
Read More
Practical Tips for Finetuning LLMs Using LoRA (Low-Rank Adaptation)

Practical Tips for Finetuning LLMs Using LoRA (Low-Rank Adaptation)

Low-rank adaptation (LoRA) is among the most widely used and effective techniques for efficiently training custom LLMs. For those interested in open-source LLMs, it's an essential technique worth familiarizing oneself with.Last month, I shared an article with several LoRA experiments, based on the open-source Lit-GPT repository that I co-maintain with my colleagues at Lightning AI. This Ahead of AI article aims to discuss the primary lessons I derived from my experiments. Additionally, I'll address some of the frequently asked questions related to the topic.  If you are interested in finetuning custom LLMs, I hope these insights will save you some…
Read More
Bringing generative artificial intelligence to space

Bringing generative artificial intelligence to space

TAMPA, Fla. — Amazon Web Services is busy positioning its cloud infrastructure business to capitalize on the promise of generative artificial intelligence for transforming space and other industries. More than 60% of the company’s space and aerospace customers are already using some form of AI in their businesses, according to AWS director of aerospace and satellite Clint Crosier, up from single digits around three years ago. Crosier predicts similar growth over the next few years in space for generative AI, which uses deep-learning models to answer questions or create content based on patterns detected in massive datasets, marking a major…
Read More
Data Machina #247

Data Machina #247

The New Breed of Open Mixture-of-Experts (MoE) Models. In a push to beat the closed-box AI models from the AI Titans, many startups and research orgs have embarked in releasing open MoE-based models. These new breed of MoE-based models introduce many clever architectural tricks, and seek to balance training cost efficiency, output quality, inference performance and much more. For an excellent introduction to MoEs, checkout this long post by the Hugging Face team: Mixture of Experts ExplainedWe’re starting to see several open MoE-based models achieving near-SOTA or SOTA performance as compared to e.g. OpenAI GPT-4 and Google Gemini 1.5 Pro.…
Read More
German AI defence giant Helsing eyes $4B valuation in Series C funding

German AI defence giant Helsing eyes $4B valuation in Series C funding

German AI defense company Helsing is reportedly in talks to secure €370 million in Series C funding, potentially valuing the company at a staggering $4 billion. The round is said to be led by US VC giant General Catalyst, according to Forbes. In September last year, the company led Helsing in securing €209 million in Series B funding, solidifying its status as Europe's most well-funded defencetech startup and among the leading AI companies in the field.  The Series C would bring its funds to almost €700 million. Founded in 2021, Helsing develops AI-based capabilities to "protect our democracies."  The company has been active in…
Read More
No widgets found. Go to Widget page and add the widget in Offcanvas Sidebar Widget Area.