Categories
Applied AI Artificial Intelligence Digital Transformation Neural Networks

Meta’s new learning algorithm can teach AI to multi-task

If you can recognize a dog by sight, then you can probably recognize a dog when it is described to you in words. Not so for today’s artificial intelligence. Deep neural networks have become very good at identifying objects in photos and conversing in natural language, but not at the same time: there are AI models that excel at one or the other, but not both. 

Part of the problem is that these models learn different skills using different techniques. This is a major obstacle for the development of more general-purpose AI, machines that can multi-task and adapt. It also means that advances in deep learning for one skill often do not transfer to others.

A team at Meta AI (previously Facebook AI Research) wants to change that. The researchers have developed a single algorithm that can be used to train a neural network to recognize images, text, or speech. The algorithm, called Data2vec, not only unifies the learning process but performs at least as well as existing techniques in all three skills. “We hope it will change the way people think about doing this type of work,” says Michael Auli, a researcher at Meta AI.

The research builds on an approach known as self-supervised learning, in which neural networks learn to spot patterns in data sets by themselves, without being guided by labeled examples. This is how large language models like GPT-3 learn from vast bodies of unlabeled text scraped from the internet, and it has driven many of the recent advances in deep learning.

Auli and his colleagues at Meta AI had been working on self-supervised learning for speech recognition. But when they looked at what other researchers were doing with self-supervised learning for images and text, they realized that they were all using different techniques to chase the same goals.

Data2vec uses two neural networks, a student and a teacher. First, the teacher network is trained on images, text, or speech in the usual way, learning an internal representation of this data that allows it to predict what it is seeing when shown new examples. When it is shown a photo of a dog, it recognizes it as a dog.

The twist is that the student network is then trained to predict the internal representations of the teacher. In other words, it is trained not to guess that it is looking at a photo of a dog when shown a dog, but to guess what the teacher sees when shown that image.

Because the student does not try to guess the actual image or sentence but, rather, the teacher’s representation of that image or sentence, the algorithm does not need to be tailored to a particular type of input.

Data2vec is part of a big trend in AI toward models that can learn to understand the world in more than one way. “It’s a clever idea,” says Ani Kembhavi at the Allen Institute for AI in Seattle, who works on vision and language. “It’s a promising advance when it comes to generalized systems for learning.”

An important caveat is that although the same learning algorithm can be used for different skills, it can only learn one skill at a time. Once it has learned to recognize images, it must start from scratch to learn to recognize speech. Giving an AI multiple skills at once is hard, but that’s something the Meta AI team wants to look at next.  

The researchers were surprised to find that their approach actually performed better than existing techniques at recognizing images and speech, and performed as well as leading language models on text understanding.

Mark Zuckerberg is already dreaming up potential metaverse applications. “This will all eventually get built into AR glasses with an AI assistant,” he posted to Facebook today. “It could help you cook dinner, noticing if you miss an ingredient, prompting you to turn down the heat, or more complex tasks.”

For Auli, the main takeaway is that researchers should step out of their silos. “Hey, you don’t need to focus on one thing,” he says. “If you have a good idea, it might actually help across the board.”

Categories
Applied AI Artificial Intelligence Machine Learning

Use Case Libraries Aim to Help Provide a Head Start With Applied AI  

By AI Trends Staff  

To accelerate the adoption of applied AI, more organizations are putting forward libraries of use cases, offering details of projects to help others get a head start.   

For example Nokia recently announced the initial deployment of multiple AI use cases delivered over the public cloud through a collaboration with Microsoft, according to a recent account in ComputerWeekly.  

Nokia, the multinational telecom company based in Finland, is suggesting that for its communication service provider (CSP) customers, AI use cases are helpful for managing the business complexity that 5G and cloud networks bring about. Nokia has integrated its AVA framework into Microsoft’s Azure public cloud digital architecture, to provide an AI-as-a-service model. 

This allows CSPs advantages in implementing AI into their networks, including faster deployment across the network and multiple clusters, with services from the Nokia security architecture available as well. Nokia suggests that AI data setup can be completed in four weeks. After that initial data setup, Nokia suggests that CSPs can deploy additional AI use cases within a week, and ramp resources up or back them off as needed across network clusters after that. The Nokia security framework on Azure is said to segregate and isolate data, to provide security equivalent to that of a private cloud.  

Friedrich Trawoeger, vice-president, cloud and cognitive services, Nokia

“CSPs are under constant pressure to reduce costs by automating business processes through AI and machine learning,” stated Friedrich Trawoeger, vice-president, cloud and cognitive services at Nokia. “To meet market demands, telcos are turning to us for telco AI-as-a-service. This launch represents an important milestone in our multicloud strategy.” Accessing the library of use cases remotely lowers costs and reduces environmental impacts, he suggested.   

Rick Lievano, CTO, worldwide telecom industry at Microsoft, stated, “Nokia AVA on Microsoft Azure infuses AI deep into the network, bringing a large library of use cases to securely streamline and optimize network operations managed by Microsoft Azure.” The offering makes the case that public clouds are able to help service providers implement AI, he suggested.  

The Australian mobile operator TPG Telecom implemented Nokia’s AVA AI on a local instance of Microsoft Azure, to help optimize network coverage, capacity, and performance. The project is said to help TPG detect network anomalies with greater accuracy and reduce radio frequency optimization cycle times by 50%.  

Declan O’Rourke, head of radio and device engineering at TPG, stated: “Nokia’s AVA AI-as-a-service utilizes artificial intelligence and analytics to help us maintain a first-class, optimized service for our subscribers, helping us to predict and deal with issues before they occur.”

AI Swedenthe Swedish National Center for Applied AI, has implemented an AI use case library to help speed adoption. “We want to accelerate the adoption of applied AI and to do so we know we need to guide businesses and organizations by showing them what is possible,” it states on the AI Sweden website. “Building an AI use case library is our way of showcasing our partners and [their] work to the rest of the world,” it says. The center offers a link to a form where anyone interested is able to add their project to the library. It asks for contact information, whether the use case is for a customer or a partner, the industry, the business function area (sales or finance), the purpose or goal, the techniques used, the sources of data and the effect of the case.  

US GSA Unit Last Year Developed a Use Case Library  

The US General Services Administration last year began developing a library of AI use cases that agencies can refer to when they start investigating the new technology. The GSA’s Technology Transformation Services (TTS) launched a community of practice to define areas where they see challenges in adopting AI, according to an account in FedScoop.  

Steve Babitch, head of AI at the GSA’s TTS, commented that the ability to search the use case library would have many advantages for project teams and could have unexpected benefits.  “Maybe there’s a component around culture and mindset change or people development,” he stated. (See Executive Interview with Steven Babitch in AI Trends, July 1, 2020.)  

Early practice areas TTS identified are acquisition, ethics, governance, tools and techniques, and possibly workforce readiness. Common early use cases across agencies include customer experience, human resources, advanced cybersecurity, and business processes.  

One example came from the Census Bureau’s Economic Indications Division (EID), where analysts developed a machine learning model to automate data coding. The division releases economic indicators for monthly retail construction data, based on a data set of all the projects in the country. They had been assigning a code to identify the type of construction taking place using a manual process.   

“It’s the perfect machine learning project. If you can automate that coding, you can speed up, and you can code more of the data,” stated Rebecca Hutchinson, big data leader at EID. “And if you can code more of the data, we can improve our data quality and increase the number of data products we’re putting out for our data users.”  

The model the EID analysts created works with about 80% accuracy, leaving 20% to be manually coded.  

Some of the analysts who helped to develop the EID ML model came out of the bureau’s data science training program, offered about two years ago to the existing workforce of statisticians and survey analysts. The program is an alternative to hiring data scientists, which is “hard,” Hutchinson stated. The training covered Python in ArcGIS and Tableau through a Coursera course. One-third of the bureau’s staff had completed training or were currently enrolled, giving them ML and web scraping skills.  

“Once you start training your staff with the skills, they are coming up with solutions,” Hutchinson stated. “It was our staff that came up with the idea to do machine learning of construction data, and we’re just seeing that more and more.” 

DataRobot Offering Use Case Library Based on its Experiences with Clients 

Another AI use case library resource is being offered by DataRobot of Boston, supplier of an enterprise AI development platform. The company built a library of about 100 use cases based on their experiences with clients in 14 industries.   

Michael Schmidt, Chief Scientist, DataRobot

“We are hyper-focused on enabling massively successful and impactful applications of AI,” stated Michael Schmidt, Chief Scientist, DataRobot, in a press release. “DataRobot Pathfinder is meant to help organizations—whether they’re customers or not—deeply understand specific applications of AI for use cases in their industry, and the right steps to create incredible value and efficiency.  

One example is an application to predict customer complaints in the airline industry. Complaints typically include flight delays, overbooking, mishandling of baggage and poor customer service. Regulations In certain geographies can result in penalties for service failures, which can be costly. Proactive responses such as emailing customers about the status of lost luggage, or a phone call to apologize for a flight delay, or financial compensation for a cancellation, can help keep customers happy. 

An AI program can provide the ability to predict when a complaint is likely, by using past complaint data. Forecasting volumes of complaints can inform call center strategy and help recommend the best service recovery solution, switching to a proactive instead of a reactive response.  

Read the source articles and information in ComputerWeekly, in FedScoop and in a press release from DataRobot.