Connect with us

Technology

Meta trains AI model to handle speech, images, text • The Register

Voice Of EU

Published

on

Researchers at Facebook parent’s Meta have trained a single AI model capable of processing speech, images, and text in the hope that these so-called multi-modal systems will power the company’s augmented reality and metaverse products.

The model, known as data2vec, can perform different tasks. Given an audio snippet, it can recognize speech. If it’s fed an image, it can classify objects. And when faced with text, it can check the grammar or analyse the writing’s tone and emotions.

AI algorithms are typically trained on one type of data, though data2vec is trained on three different modalities. It still, however, processes each form, whether its speech, images, and text, separately.

Meta believes these multi-modal models will help computers be more adaptable to blend physical and digital environments into one. “People experience the world through a combination of sight, sound and words, and systems like this could one day understand the world the way we do,” Meta CEO Mark Zuckerberg said in a statement to El Reg.

“This will all eventually get built into AR glasses with an AI assistant so, for example, it could help you cook dinner, noticing if you miss an ingredient, prompting you to turn down the heat, or more complex tasks.”

Data2vec is a transformer-based neural network and uses self-supervised learning to learn common patterns in audio, computer vision, and natural language processing. The model learns to operate with different types of data by learning how to predict how the representation of data it’s given; it knows it has to guess the next group of pixels when given an image, or the next speech utterance in audio, or fill in the words in a sentence.

The researchers used a mix of 16 Nvidia V100 and A100 GPUs to train data2vec on 960 hours of speech audio, millions of words from books and Wikipedia pages, and images from ImageNet-1K.

“We train separate models for each modality but the process through which the models learn is identical,” Alexei Baevski, a research engineer at Meta AI told The Register.

“We hope that it will enable future work to build high performing self-supervised models that combine modalities and are more effective than specialized models. Different modalities can add additional information to the same piece of content – for example body language from video, prosodic information from audio, and text can combine into a richer representation of a dialog. The algorithms that currently try to combine multi-modal information exist but they do not yet perform well enough to replace specialized algorithms and we hope our work will help change that.”

Baevski said in the future multi-modal systems could incorporate a larger range of data to model concepts like smell, 3D objects, or videos. He referred back to the idea of AR glasses helping wearers cook.

“Imagine having a model that has been trained on recordings of thousands of hours of cooking activity from various restaurants and chefs. Then, when you are cooking in a kitchen wearing your AR glasses that have access to this model, it’s able to overlay visual cues for what you need to do next, point out potential mistakes, or explain how adding a particular ingredient will affect the taste of your dish,” he told us.

Previous research on multi-modal systems have shown they can be prone to easy adversarial attacks. OpenAI’s CLIP model, for example, trained on images and text will identify an image of an apple incorrectly as an iPod if the word “iPod” are in the picture. It’s unclear, however, if data2vec suffers from similar weaknesses.

“We have not specifically analyzed how our models will react to adversarial examples but since our current models are trained separately for each modality, we believe that existing research on adversarial attack analysis for each modality would be applicable to our work as well,” Baevski said.

“In the future, we hope to use our work to enable high performance algorithms that combine modalities in one model and we plan to study how susceptible they are to adversarial attacks.”

When the researchers tested data2vec, it outperformed some top models that had been trained on a specific data type only on different types of tasks. The preliminary results are described in a paper [PDF], and the code has been published on GitHub.

“Data2vec demonstrates that the same self-supervised algorithm can work well in different modalities — and often better than the best existing algorithms,” the researchers explained in a blog post this week.

“This paves the way for more general self-supervised learning and brings us closer to a world where AI might use videos, articles, and audio recordings to learn about complicated subjects, such as the game of soccer or different ways to bake bread. We also hope data2vec will bring us closer to a world where computers need very little labeled data in order to accomplish tasks.” ®

Source link

Technology

Brace for a shock: cost-of-living crisis drives up price of electric car charging | Electric, hybrid and low-emission cars

Voice Of EU

Published

on

While petrol price rises may have made the headlines, the energy crisis has also been hitting owners of electric cars in the pocket. The cost of charging at home has risen by 43% for some drivers, while the already higher cost of on-the-road recharges has gone up 25%.

As energy prices are forced up due to rising costs for suppliers, specialist charging deals for drivers have become more scarce. And now there are suggestions that people may put off the purchase of an electric car as the cost-of-living crisis takes hold.

Although demand for vehicles is high, a new report to be released this week from Volkswagen Financial Services suggests that fewer people might commit to buying electric vehicles (EVs) as belts tighten and the cost of energy increases.

“The cost-of-living squeeze will probably mean some potential EV purchasers may not commit to a switch this year, particularly as such vehicles are perceived to be more expensive in relative terms when compared to combustion engine alternatives,” says the report.

Home charging

Electric car owners who are charging their vehicle at home will usually find the most cost-efficient option is one of the specialist tariffs on offer. “Two-rate” tariffs offer one price for electricity used during the day and another for night-time use. When prices are much lower you can top up your battery cheaply.

For example, comparison site Love My EV lists the rates for EDF’s GoElectric 35 as 44.69p per kilowatt hour (p/kWh) during the day and 4.5p/kWh at night. The Octopus Go tariff costs 35.04p/kWh during the day and 7.5p/kWh at night. Both figures are based on supplying a home in south Wales.

Three electric cars charging at a roadside station with an attractive yellow zig-zag canopy sheltering the chargers
A public charging station in Sunderland: many electric vehicle owners cannot charge at home and must pay on-the-road rates. Photograph: Christopher Thomond/The Guardian

Since energy prices have increased, the number of specialist deals on the market has dropped, says Laura Thomson, co-founder of Love My EV. While they are usually the best deals for drivers who charge overnight, the day rate and standing charge can be expensive, which consumers need to take into account when working out what is best for their situation.

“For most people who have an EV to charge at home, it does make sense, but there is a high standing charge and a high day rate to factor in,” says Thomson. If you use a lot of electricity during the day, this may not be your best option.

The site has a comparison tool for tariffs. Beware of promises of “free miles” within tariffs as these savings may be outweighed by higher charges, it says.

The rising price of EV tariffs means drivers now face paying 43% more than a year ago. This amounts to a rise of about £75 a year for an average vehicle such as a Nissan Leaf or a Renault Zoe, says Ben Nelmes of transport research company New AutoMotive.

In 2021, the cost of recharging an EV that covered 7,400 miles a year – the average mileage – and was recharged mostly at night was £174. This was based on an overnight rate of 4p/kWh and a day rate of 18p/kWh. By last month, this same charging practice cost £249 a year, based on the best prices then available – 5p/kWh at night and 28p/kWh during the day.

“Someone driving a bigger EV, such as a Kia e-Niro or Tesla, will find that this underestimates what they’ll be paying. Similarly, someone in a Smart car will find they spend a bit less than this,” says Nelmes.

On the road

Rising costs have also become apparent at public chargers. Instavolt, which operates a charging network across Britain, has increased its prices twice so far this year, first from 45p/kWh to 50p/kWh and then to 57p/kWh. Ubitricity, one of London’s largest charging networks, increased prices from 24p/kWh to 32p/kWh last month.

Data company Zap Map, which maps public charge points, found that, on average, charging costs increased from 24p/kWh in December to 30p/kWh in February for slow and fast chargers, and from 35p/kWh to 44p/kWh for rapid and ultra-rapid chargers.

“The price of charging your EV on the public network, or at home, has risen substantially over the past few months with the general increase in electricity prices,” says Melanie Shufflebotham from Zap Map.

There are 460,000 EVs currently in the UK, according to the Volkswagen Financial Service report, and just 300,000 home charger points installed. Those who don’t have a home charger end up paying more, according to Keith Brown of Paythru, a payments technology company. “One of the big inequities of the emerging EV charging market is the price ‘premium’ electric vehicle drivers pay if they don’t or can’t have a home charge point,” he says. “Domestic supply is taxed at a VAT rate of 5% whereas public charge-point supply is taxed at a VAT rate of 20%.”

Shufflebotham has called for the rates to be made equal. “Equalising the VAT rate for both public and home charging would be a great example of levelling up, and encourage more people to make the transition to electric vehicles,” she says.

The advantages

Despite increasing prices, EV drivers still face much lower bills than those with petrol or diesel cars, using figures based on the same annual mileage for all types of vehicle.

Nelmes says that while the rises in the costs of EV charging at home are high, they are dwarfed by the costs of filling a car with fuel.

“We estimate the average UK motorist would spend £1,028 per year on petrol and £987 per year on diesel. That’s up from £796 a year on petrol and £747 a year on diesel a year ago,” he says. “That means that the fuel cost savings available to petrol and diesel drivers who switch to EVs this year are £779 for petrol drivers and £738 for diesel drivers.”

Case study: positives and negatives

Having bought a Nissan Leaf in the last few weeks, Philip Ingram looks back at the deals that were available last year with some annoyance.

He currently pays a flat rate throughout the day of 28.45p/kWh with British Gas, the best tariff available to him at home in Bordon, Hampshire. Last year, he could have taken advantage of deals of 5p/kWh overnight, he says. While there are deals with good night-time rates, now their high day rates mean they do not suit the family budget.

The annoyance is tempered by the savings from moving from a diesel VW Golf to an EV.

Ingram, who runs a cotton company called LittleLeaf Organic, used to pay nearly £90 to fill up with diesel but gets the same mileage for £20 of charging. This has to be balanced against the cost of the car: £24,000. “I wish we had done it a long time ago,” he says, “but the reason that we have been slower is … capital costs. Several times I have said to [my wife] Lisa the running costs are unbelievable, but then you look at the cost of buying this car, [which] is enormous.”

Source link

Continue Reading

Technology

Web ad firms scrape email addresses before you know it • The Register

Voice Of EU

Published

on

Tracking, marketing, and analytics firms have been exfiltrating the email addresses of internet users from web forms prior to submission and without user consent, according to security researchers.

Some of these firms are said to have also inadvertently grabbed passwords from these forms.

In a research paper scheduled to appear at the Usenix ’22 security conference later this year, authors Asuman Senol (imec-COSIC, KU Leuven), Gunes Acar (Radboud University), Mathias Humbert (University of Lausanne) and Frederik Zuiderveen Borgesius, (Radboud University) describe how they measured data handling in web forms on the top 100,000 websites, as ranked by research site Tranco.

The boffins created their own software to measure email and password data gathering from web forms – structured web input boxes through which site visitors can enter data and submit it to a local or remote application.

Providing information through a web form by pressing the submit button generally indicates the user has consented to provide that information for a specific purpose. But web pages, because they run JavaScript code, can be programmed to respond to events prior to a user pressing a form’s submit button.

And many companies involved in data gathering and advertising appear to believe that they’re entitled to grab the information website visitors enter into forms with scripts before the submit button has been pressed.

“Our analyses show that users’ email addresses are exfiltrated to tracking, marketing and analytics domains before form submission and without giving consent on 1,844 websites in the EU crawl and 2,950 websites in the US crawl,” the researchers state in their paper, noting that the addresses may be unencoded, encoded, compressed, or hashed depending on the vendor involved.

Most of the email addresses grabbed were sent to known tracking domains, though the boffins say they identified 41 tracking domains that are not found on any of the popular blocklists.

“Furthermore, we find incidental password collection on 52 websites by third-party session replay scripts,” the researchers say.

Replay scripts are designed to record keystrokes, mouse movements, scrolling behavior, other forms of interaction, and webpage contents in order to send that data to marketing firms for analysis. In an adversarial context, they’d be called keyloggers or malware; but in the context of advertising, somehow it’s just session-replay scripts.

Gunes Acar, one of the report co-authors, was also the co-author of a similar research project in 2017 that looked at data gathering by session-replay companies Yandex, FullStory, Hotjar, UserReplay, Smartlook, Clicktale, and SessionCam.

Evidently, not much has changed since then, except perhaps that email addresses have become more desirable as unique identifiers now that privacy-oriented browsers like Brave, Firefox, and Safari are taking more steps to block cookies and tracking scripts.

Email addresses, the researchers observe, represent a cookie replacement because they’re unique, persistent, and can be used to track people across applications, platforms, and even offline interactions that may be tied to an email address like loyalty card transactions.

The website categories with the most leaking forms include: Fashion/Beauty (11.1 per cent, EU; 19 per cent US); Online Shopping (9.4 per cent EU; 15.1 per cent US); and General News (6.6 per cent EU; 10.2 per cent US).

Websites categorized as Pornography had the best privacy when it comes to surreptitious form data harvesting.

“A somehow surprising result was the following: despite filling email fields on hundreds of websites categorized as Pornography, we have not a single email leak,” the researchers say, noting that previous studies of adult-oriented websites have relatively fewer third-party trackers than similarly popular general interest websites.

Those pesky regulations

The report authors say that EU websites practicing email exfiltration may be in violation of at least three GDPR requirements: transparency, purpose limitation, and prior consent. Firms found to be violating these rules can be fined up to $20m euros or 4 per cent of annual revenue, per Article 83(5).

The US doesn’t have a federal data privacy law, though it’s conceivable one of the handful of US states with applicable privacy rules could take action against pre-submission form harvesting. But given the toothlessness of US privacy regulation over the past decade, don’t expect much.

The authors say they attempted to contact 58 first-parties and 28 third-parties with GDPR requests. They report receiving 30 responses from the first-parties, which varied from surprise and remediation to justifications of one sort or another.

“fivethirtyeight.com (via Walt Disney’s DPO), trello.com (Atlassian), lever.co, branch.io and cision.com were among the websites that said they had not been aware of the email collection prior to form submission on their websites and removed the behavior,” the report says.

Marriott, meanwhile, said the information collected by digital analytics firm Glassbox helps with customer care, technical support, and fraud prevention.

Third-parties Taboola, Zoominfo, and ActiveProspect defended their data collection practices.

Facebook, aka Meta, is among the third-parties involved in this. The researchers say that email addresses or their hashes were spotted being sent to facebook.com from 21 different websites in the EU.

“On 17 of these, Facebook Pixel’s Automatic Advanced Matching feature was responsible for sending the SHA-256 of the email address in a SubscribedButtonClick event, despite not clicking any submit button,” the report says.

Advanced Matching – called out recently for harvesting student loan data – is designed to collect hashed customer data, such as email addresses, phone numbers, and names from checkout, sign-in, and registration forms. The researchers speculate that on these sites, Facebook’s script treats clicks on non-submit buttons as a click event for the submit button.

Facebook did not respond to a request for comment.

The report concludes that browser vendors, regulators, and privacy tool makers need to deal with this issue because it isn’t going away. “Based on our findings, users should assume that the personal information they enter into web forms may be collected by trackers – even if the form is never submitted,” the report concludes. ®

Source link

Continue Reading

Technology

VC funding in Ireland rose in Q1, but not for deals under €10m

Voice Of EU

Published

on

A William Fry-commissioned report has found that funding deals under €10m have taken a big hit in the first three months of 2022.

Venture capital funding into Irish tech businesses was up by more than 50pc in the first quarter of this year, but there’s an unfortunate and potentially troubling caveat to that.

The Irish Venture Capital Association (IVCA) has published today (15 May) its latest report on VC funding into tech start-ups and SMEs in Ireland, which found that the investments increased by 52pc to €379.7m in the first three months of 2022, compared to the same period last year.

Future Human

But the report, commissioned by Dublin law firm William Fry, also found that VC funding in deals valued less than €10m have taken a hit.

IVCA chair Nicola McClafferty said that the headline figure of a funding boost conceals a “potentially worrying fall” of 30 to 50pc across all categories of deals under €10m – including seed funding.

“All the growth came from eight deals worth over €10m each, including three over €30m. While the momentum carried over from last year has continued for more established companies raising large rounds, some of that impetus seems to have stalled for earlier stage companies.”

Even the total number of deals overall fell by almost a third to 50 from 74 in the same period last year.

McClafferty said that this could be related to international trends affecting the business world right now, such as Russia’s invasion of Ukraine.

“While challenging market conditions may continue, we also know that many great companies are started and built in times of downturn, so we await with interest the data in the coming quarters,” she added.

Deals in the €5m to €10m range fell in value by more than half, while those in the €1m to €5m range also halved from €70.3m last year to €34.5m in Q1 2022. The value of deals below €1m dropped by 31pc to €8.9m.

Seed funding also took a hit, falling by nearly 40pc to €22.3m from €36.5m last year.

Nearly four-fifths of all funding came from overseas sources, according to IVCA director-general Sarah-Jane Larkin.

“While this is to be welcomed and emphasises the quality of Irish tech firms and their appeal to international investors, we have expressed concern before about where any shortfall would be made up if the global economy contracts,” she said.

Wayflyer, Ireland’s latest tech unicorn, led the way in terms of total value of funding received with a $150m in Series B funding valuing the start-up at $1.6bn. Flipdish, another Irish tech start-up that became a unicorn this year, raised $100m reaching a $1.25bn valuation.

10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.

Source link

Continue Reading

Trending

Subscribe To Our Newsletter

Join our mailing list to receive the latest news and updates 
directly on your inbox.

You have Successfully Subscribed!