NVIDIA announced in its speech AI summit that it would collaborate with Mozilla Common Voice for its new speech AI. Both NVIDIA and Mozilla Common Voice aim to speed up the growth of automatic speech recognition models.
Mozilla Common Voice is an open-source platform with multi-language datasets of voices that users can use to train any speech-enabled application. It has datasets for 34 languages, including Hindi, English, Bengali, Marathi, Tamil, and more.
NVIDIA realized that standard voice assistants such as Amazon Alexa and Google have very few of the world’s spoken languages. To solve this problem, NVIDIA and Mozilla Common Voice decided to improve linguistic inclusion in speech AI.
In the same speech AI summit, Caroline de Brito Gottlieb, product manager at NVIDIA, said that Demographic diversity is the key to capturing language diversity. She also stated that many factors, like underserved dialects, pidgins, and accents, can impact speech variation. With the NVIDIA-Mozilla Common Voice partnership, NVIDIA can create a dataset ecosystem to build speech datasets and models for any language.
NVIDIA has been developing speech AI for many use cases, like artificial speech translation, automatic speech recognition, and text-to-speech. NVIDIA’s Riva is a GPU-accelerated speech AI SDK used to build and deploy fully customizable real-time AI pipelines.
Data-centric AI is an emerging class of AI that focuses on “data” rather than the model. Generally, machine learning (ML) techniques are model-centric; they attain a static environment within which the model performs. However, most AI applications in the real world cannot always function in a static environment because several processes incorporating a machine learning approach require different processing and monitoring capabilities. To make AI applications more efficient and standardized, models are shifting to a data-centric approach or a combination of both. The new data-centric approach focuses on studying, analyzing, and utilizing data for decision-making.
Andrew Ng, a deep-learning pioneer, and founder of Landing AI has become a vocalist for data-centric AI as he believes everything comes down to data. If data is carefully prepared, organizations can accomplish the same goals with much lesser of it. To reach this stage, all organizations must shift to a data-centric approach to reap the maximum benefits of using artificial intelligence.
This shift in approach has pushed for “Active Learning (AL)” to reduce manual efforts of sampling and labeling data in ML models. Active Learning shortlists the most representative data samples for training and sends them for labeling. Only the selected sub-datasets are fed into the model to obtain more competitive results, save labeling time, and reduce training costs. While this approach saves manual data handling time, users must build an extensive backend to run those active learning pipelines. Consequently, significant engineering and coding work make active learning application challenging.
To overcome the issue, the newly proposed system at the National University of Singapore, named Active-Learning-as-a-Service (ALaaS), runs multiple strategies on datasets and performs the desired tasks by building pipelines. The server-client architecture, data manager, and AL strategy zoo are the three major components of this framework.
The system adopts a “server-client” architecture to perform scheduled jobs, making it compatible with individual devices and clouds. This architecture abstracts all necessary algorithms into web-based services users can directly use. Users only need to follow the suggested guidelines while creating a configuration file with basic settings like dataset path and desired techniques. Users can then initiate the client and the server with only a few lines of code (LoCs).
Once a user uploads the dataset and initiates the server, the data manager becomes responsible for it. The manager stores the metadata and indexes the samples to avoid redundant data movements.
Finally, the AL strategy zooabstracts the desired strategies, like Bayesian, density-driven, batch selection, etc.
Besides the three main components, others, like the model repositoryand serving engine, help automate the AL application by enabling connections with public hubs like HuggingFace, TorchHub, etc., and calling other ML serving backends for inference.
With its smartly designed architecture and components, ALaaS promises three key improvements: efficiency, modularity, and accessibility.
ALaaS makes it much more convenient to leverage active learning by provisioning optimization technologies, including pipeline generation, ML backend adoption, etc. As active learning mainly faces large-scale datasets while employing multiple computational deep learning (DL) models, dataset processing, application development, and ML backend adoption are highly crucial for efficiency.
ALaaS leverages a user-friendly experience by implementing containerized active learning services to ensure that even non-technical users can conveniently use it without getting into code details, making it highly accessible.
Active learning is advancing rapidly, mainly due to the development of deep learning techniques. Making active learning more accessible should not prevent professionals from using it for more complex projects. Due to the very modular nature of ALaaS, professionals may quickly prototype, extend, and deploy cutting-edge, state-of-the-art active learning approaches.
Such an MLOps system, as described in “Active-Learning-as-a-Service: An Efficient MLOps System for Data-Centric AI,” for leveraging data-centric AI is a significant advancement in Machine-Learning-as-a-Service. More crucially, ALaaS integrates batching, cache, and stage-level parallelism (a pipeline approach) to increase the effectiveness of active learning operations. Results from the study also show that active learning offers low latency and higher throughput than running individual jobs.
Meta has announced that it is testing, minting, and selling NFTs (or, as Meta calls it, digital collectibles) on Instagram, with a select set of creators in the United States being the first to have access to the feature.
The first group of chosen creators consists of Amber Vittoria, Dave Krugman, Refik Anadol, Isaac “Drift” Wright, Eric Rubens, Jason Seife, Vinnie Hager, Sara Baumann, Olive Allen, and Ilse Valfre. Creators will now have access to a toolbox that will enable them to produce, promote, and sell digital collectibles. Creators can showcase NFTs issued on the Polygon, Flow, Solana, and Ethereum blockchains. Similar to what Twitter does for its NFT profile image feature, Instagram will also be adding some metadata from OpenSea to the display.
Image Credit: Meta
According to Meta, there will be no costs for posting and sharing digital collectibles on Facebook or Instagram, and there won’t be any further fees for selling digital collectibles until at least 2024. Moreover, it promised that for digital collectibles purchased on Instagram at launch, neither artists nor collectors would have to pay gas costs. However, it clarified that digital collectable transactions made within the Instagram app for the Android and iOS operating systems are liable for applicable app store fees.
Top: the process of making and selling NFTs on Instagram. Bottom: the process of buying an NFT on Instagram. Image Credit: Meta
It is reported that NFT creators will also be to choose their royalty portion, which will likely range from 5% to 25%. Then, creators can connect their bank account or Paypal account to get payment.
After the NFT announcements, Meta added Instagram is extending subscription access to all U.S.-based qualified creators. With a small number of producers, the social network started experimenting with subscriptions in January. The function, which was first seen on the App Store in November 2021, allows creators to charge their followers for exclusive Instagram Live videos and Stories. Subscribers are also issued a unique badge that helps them stand out in the comments area and in the inboxes of the creators.
Meanwhile, Meta is extending its professional-mode profile setting on Facebook to all creators. The professional mode is intended for usage by creators who want to utilize social networking site to monetize their fan bases. In December 2021, Facebook began testing professional mode with a small group of creators; it is now accessible to everyone on the network.
Meta also revealed that it would be launching gifts on Instagram, beginning with reels, to enable artists a new option to get paid by their followers. Fans may send gifts on reels by purchasing Stars on Instagram. Stars are virtual commodities that allow fans to show their support for their favorite creators during Facebook videos and live streams. This feature is now being tested by Meta with a select selection of American creators, and it will soon be made available to additional creators.
Further, Meta said that Facebook is extending Stars’ user base, enabling creators to get paid directly by viewers of Reels, live events, and recorded videos. Facebook will also begin testing automated creator onboarding, which will result in the ability to send stars showing up on their content without any hassles. Besides this, Stars are being added to non-video material on Facebook, such as images and textual posts.
The San Francisco-based lab, OpenAI, has launched a new program called Converge to provide early-stage AI startups with a capital of $1 million each and access to OpenAI resources.
The ten founders chosen for Converge will receive $1 million each, along with five weeks of office hours, events, and workshops with OpenAI staff. They will also receive early access to OpenAI models and programs tailored for AI companies. The deadline for application is November 25. However, OpenAI notes that it will continue to evaluate applications after the deadline for future cohorts.
When OpenAI first announced the OpenAI Startup Fund, it mentioned that the recipients of cash from the fund will also receive access to Azure resources from Microsoft. It is unclear if the same benefit will be available for Converge participants.
With Converge, OpenAI is looking to cash in on the increasingly lucrative AI industry. The Information reports that OpenAI, which is reportedly in talks to raise cash from Microsoft at a nearly $20 billion valuation, has agreed to lead the financing of Descript, an AI-powered audio and video editing app, at a valuation of around $550 million.
Edtech platform Byju’s, decided to revoke its plan to shut its operations in Thiruvananthapuram on 2 November after CEO Byju Raveendran met with Pinarayi Vijayan, Kerala’s Chief Minister.
Apart from revoking the plan to layoff and relocate 140 employees to one of its offices in Kerala, CEO Raveendran announced BYJU’s plans to hire 600 people in Thiruvananthapuram. The company has almost 3,000 people employed in Kerala.
The following update arrives as Raveendran, had written an emotional mail to the 2,500 employees on 31 October saying he is planning to layoff employees to ensure capital-efficient growth and sustainability due to adverse macroeconomic factors.
On 25 October, 170 Byju’s employees reached out to the Kerala labor commissioner K Vasuki claiming that a verbal request for a forced resignation by the company. At the same time, several media reports stated that the company had offered the laid-off employees to relocate to Kochi or Bengaluru.
Commenting on the layoff, Raveendran said he is genuinely sorry to those who will have to leave the company, adding that sackings break his heart too. He also sought forgiveness if the process was not smooth for the employees.
Raveendran said some business decisions need to be taken to protect the well-being of the larger organization, maintaining the decision did not reflect their (employees) performances and assured them support in their transition.
Elon Musk, now the sole director and CEO of Twitter, has authorized employees from other companies that he owns to work at the social media giant. These include 50 employees from Tesla, mainly from its Autopilot team, two from Boring Company, and one from Neuralink.
The employees from other companies that Musk has involved into Twitter include those that he ‘trusts’, like Tesla’s senior director of software engineering Maha Virduhagiri, director of software development Ashok Elluswamy, director of Autopilot/TeslaBot engineering Milan Kovac, and others.
Shortly after taking over, Musk fired the company’s top bosses – CEO Parag Agrawal, CFO Ned Segal, and the legal and policy head, Vijaya Gadde and dissolved the company’s board of directors.
He has now announced that Twitter will charge $8 per month for the blue verification tick that authenticates a Twitter account. Musk changed the “current lords and peasants system” for who does or does not have a blue checkmark.
“Power to the people! Blue for USD 8 per month,” he had tweeted, adding that the price is adjusted by country proportionate to purchasing power parity. Twitter already has a subscription service called Twitter Blue, which launched in June last year and offers access to features such as an option to edit tweets.
Larsen and Turbo Technology (L&T) Technology Services announces to open its new Engineering, Research, and Development (ER&D) center in Toronto, Canada. This is the third global design center of L&T after having two ERD centers in France and Polland.
This new center will address Canada-based clients to develop solutions for digital products. It will be the nearshore site for many American-based customers, and plans to hire more than 100 engineers in the next 12-24 months.
Amit Chanda, Chief Executive Officer of L&T Technology Services, stated that through this new center in Toronto, customers from Canada and North America would use the company’s cutting-edge technology and digital products.
This center can focus on developing digital solutions for the transportation sectors like railway engineering. Besides railway engineering, the new L&T center can also build applications for digital asset management, sensors, advanced mobility solutions, communications systems, and digital flyboard.
The inauguration of this new ER&D center in Toronto is a significant step towards further strengthening the two countries’ relationship and encouraging the Canada-India economic corridor.
Ruben Villegas and a few other researchers at Google unveil Phenaki, a system that generates videos from story-like descriptions given as text prompts. There are only a few datasets that can be used for text-to-video generation, but there are many text-to-image pairs available, using which Google has also developed text-to-image frameworks like Imagen.
Now, text-to-video generator Phenaki generates short videos by using images as single-frame videos and clubbing them together with a dataset of short videos having captions.
1/ From today's AI@ event: we announced our Imagen text-to-image model is coming soon to AI Test Kitchen. And for the 1st time, we shared an AI-generated super-resolution video using Phenaki to generate long, coherent videos from text prompts and Imagen Video to increase quality. pic.twitter.com/WofU5J5eZV
Phenaki works using some main components. It uses an encoder for video embedding, a language model for text embedding, a MaskGIT bidirectional transformer, and a decoder.
The system uses a “videos less than three seconds long” dataset to train the C-ViViT encoder/decoder to generate embeddings. The encoder is trained to generate non-overlapping patches as vectors by splitting frames. The decoder is trained to convert embeddings into pixels.
Phenaki uses the t5x language model to produce text embedding. MaskGIT generates the masked embeddings at inference using a set of masked video embeddings and text embeddings and then re-masks a portion of them to be generated in subsequent iterations.
To create minute-long videos, the authors repeatedly combined MaskGIT and C-ViViT. They first created a short film from a single sentence, after which they encoded the final k frames. They combined the text after the video embeddings to create more video frames. Unlike Make-A-Video, which uses several diffusion models to generate short videos and then upscale its resolution, Phenaki bootstraps its own frames to enhance throughput and narrative complexity.
Researchers from IIIT Delhi used the conventional Baum-Welch algorithm and deep learning to devise novel AI-driven approaches to predict crypto pricing and other financial parameters. As cryptocurrencies are not pegged against standard parameters or products, speculating their prices is challenging.
Shalini Sharma, a Ph.D. scholar from IIIT-Delhi, and her supervisor Dr. Angshul Majumdar devised two predominant approaches to predict financial parameters like crypto prices. Elsevier Information Sciences has recently validated the work and declared it to be “very precise” in predicting future prices.
The first approach builds on the Baum-Welch framework, a particular case of expectation-maximization used in a Hidden Markov Model (HMM). Using this framework, users can predict not only the prices but also the prediction uncertainty. This strategy calls for understanding the underlying factors influencing price swings, which is only sometimes possible with cryptocurrency.
The research also shows that the uncertainty estimates received from the first approach are correlated with historical CVI values. CVI, or crypto volatility index, shows how crypto prices react to fluctuations over time.
The second approach is driven by DL, as it does not require prior knowledge about underlying factors. This approach can predict crypto prices but cannot give uncertainty, making it ineffective in interpretability aspects.
The work is a significant step forward to aid crypto enthusiasts in having confidence in mining cryptocurrencies.
Hong Kong is all set to relax its strict crypto regulation with a plan to allow retail cryptocurrency trading. According to the report, a mandatory licensing regime for crypto platforms that enables retail crypto trading is set to be enforced in March 2023.
It said that Hong Kong is planning to legalize retail trading for cryptocurrency starting in March after years of skepticism, contrasting with ban in mainland China’s.
Moreover, regulators are also planning to allow retail exchanges to list prominent cryptocurrencies, like bitcoin (BTC) and ether (ETH). The listing rules will likely include criteria such as the token’s inclusion, market value, liquidity in third-party crypto indexes.
Executive director of crypto firm BC Technology Group, Gary Tiu, commented that introducing mandatory licensing in Hong Kong is one of the essential things regulators must do. They cannot forever effectively close the needs of retail investors, he added.
Executive president of Hashkey, digital asset financial services group, Michel Lee, explained that Hong Kong has been toiling to create an all-encompassing crypto regime, citing tokenized bonds and stocks as a potentially more critical segment in the future. “Just trading digital assets on their own is not the goal. The goal is to grow the ecosystem,” he said.
The Securities and Futures Commission (SFC), Hong Kong’s top financial regulator, introduceD a voluntary licensing regime in the year 2018. It limited crypto trading platforms to clients with portfolios of minimum HK$8 million ($1 million). However, the strict regulation turned away several crypto businesses, and only two firms, i.e., BC Technology Group and Hashkey, were approved.
