Connect with us

Cars

Google ponders the shortcomings of machine learning

Published

on

Critics of the current mode of artificial intelligence technology have grown louder in the last couple of years, and this week, Google, one of the biggest commercial beneficiaries of the current vogue, offered a response, if, perhaps, not an answer, to the critics.

In a paper published by the Google Brain and the Deep Mind units of Google, researchers address shortcomings of the field and offer some techniques they hope will bring machine learning farther along the path to what would be “artificial general intelligence,” something more like human reasoning.

The research acknowledges that current “deep learning” approaches to AI have failed to achieve the ability to even approach human cognitive skills. Without dumping all that’s been achieved with things such as “convolutional neural networks,” or CNNs, the shining success of machine learning, they propose ways to impart broader reasoning skills.

Also: Google Brain, Microsoft plumb the mysteries of networks with AI

The paper, “Relational inductive biases, deep learning, and graph networks,” posted on the arXiv pre-print service, is authored by Peter W. Battaglia of Google’s DeepMind unit, along with colleagues from Google Brain, MIT, and the University of Edinburgh. It proposes the use of network “graphs” as a means to better generalize from one instance of a problem to another.

Battaglia and colleagues, calling their work “part position paper, part review, and part unification,” observe that AI “has undergone a renaissance recently,” thanks to “cheap data and cheap compute resources.”

However, “many defining characteristics of human intelligence, which developed under much different pressures, remain out of reach for current approaches,” especially “generalizing beyond one’s experiences.”

Hence, “A vast gap between human and machine intelligence remains, especially with respect to efficient, generalizable learning.”

The authors cite some prominent critics of AI, such as NYU professor Gary Marcus.

In response, they argue for “blending powerful deep learning approaches with structured representations,” and their solution is something called a “graph network.” These are models of collections of objects, or entities, whose relationships are explicitly mapped out as “edges” connecting the objects.

“Human cognition makes the strong assumption that the world is composed of objects and relations,” they write, “and because GNs [graph networks] make a similar assumption, their behavior tends to be more interpretable.”

Also: Google Next 2018: A deeper dive on AI and machine learning advances

The paper explicitly draws upon work for more than a decade now on “graph neural networks.” It also echoes some of the recent interest by the Google Brain folks in using neural nets to figure out network structure.

But unlike that prior work, the authors make the surprising assertion that their work doesn’t need to use neural networks, per se.

Rather, modeling the relationships of objects is something that not only spans all the various machine learning models — CNNs, recurrent neural networks (RNNs), long-short-term memory (LSTM) systems, etc. — but also other approaches that are not neural nets, such as set theory.

The Google AI researchers reason that many things one would like to be able to reason about broadly — particles, sentences, objects in an image — come down to graphs of relationships among entities.


Google Brain, Deep Mind, MIT, University of Edinburgh.

The idea is that graph networks are bigger than any one machine-learning approach. Graphs bring an ability to generalize about structure that the individual neural nets don’t have.

The authors write, “Graphs, generally, are a representation which supports arbitrary (pairwise) relational structure, and computations over graphs afford a strong relational inductive bias beyond that which convolutional and recurrent layers can provide.”

A benefit of the graphs would also appear to be that they’re potentially more “sample efficient,” meaning, they don’t require as much raw data as strict neural net approaches.

To let you try it out at home, the authors this week offered up a software toolkit for graph networks, to be used with Google’s TensorFlow AI framework, posted on Github.

Also: Google preps TPU 3.0 for AI, machine learning, model training

Lest you think the authors think they’ve got it all figured out, the paper lists some lingering shortcomings. Battaglia & Co. pose the big question, “Where do the graphs come from that graph networks operate over?”

Deep learning, they note, just absorbs lots of unstructured data, such as raw pixel information. That data may not correspond to any particular entities in the world. So they conclude that it’s going to be an “exciting challenge” to find a method that “can reliably extract discrete entities from sensory data.”

They also concede that graphs are not able to express everything: “notions like recursion, control flow, and conditional iteration are not straightforward to represent with graphs, and, minimally, require additional assumptions.”

Other structural forms might be needed, such as, perhaps, imitations of computer-based structures, including “registers, memory I/O controllers, stacks, queues” and others.

Previous and related coverage:

What is AI? Everything you need to know

An executive guide to artificial intelligence, from machine learning and general AI to neural networks.

What is deep learning? Everything you need to know

The lowdown on deep learning: from how it relates to the wider field of machine learning through to how to get started with it.

What is machine learning? Everything you need to know

This guide explains what machine learning is, how it is related to artificial intelligence, how it works and why it matters.

What is cloud computing? Everything you need to know about

An introduction to cloud computing right from the basics up to IaaS and PaaS, hybrid, public, and private cloud.

Related stories:



Source link

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Cars

Rivian EV configurator opens to all – R1S and R1T Launch Edition sold out

Published

on

Rivian has thrown open access to its online configurator, meaning you no longer need to have a reservation for the R1T or R1S in order to customize your perfect electric truck. Set to begin manufacturing and deliveries next year, the two EVs share the same platform – the R1T having a pickup body, while the R1S is a full-size SUV – though are likely to appeal to different markets.

We saw the first results of the configurator last week, when Rivian granted access to those who had paid the $1,000 deposit to stake a place in line. In the process it confirmed some of the options that buyers will be able to pick from, including multiple paint finishes, different interior trims, and some of the more unusual accessories.

The R1T, for example, can be equipped with a slide-out mini kitchen for camping. That has a sink – with a water tank and pump that’s powered by the trunk’s own battery – along with an induction stove for cooking. Rivian even has a custom set of prep and cookware from Snow Peak to go with it.

Arguably more useful every day, meanwhile, is the Max Pack battery. Offered only on the R1T pickup, it’s not inexpensive at $10,000, but it boosts the estimated range from the standard 300+ miles to 400+ miles. Final EPA-certified range is unlikely to be confirmed until next year, closer to the R1T’s summer release.

While it’s nice to be able to tinker with the configurator, there’s also some bad news if you were hoping for a R1S or R1T Launch Edition. Reservations for that special trim are now full, Rivian has confirmed, closing the order books on the very first examples of the two EVs. Priced at $75,000 for the pickup, and $77,500 for the SUV, the Launch Edition is prety much a maxed-out example of each, and offers exclusive options like Launch Green paintwork.

It means that, if you didn’t get your order in already, you’ve some wait ahead of you. The two mainstream trims for both EVs – the entry-level Explore and the better-equipped Adventure – are both available to order, but deliveries aren’t expected to begin until January 2022.

Before then, we may have heard more about some of Rivian’s upcoming competition. Ford’s all-electric F-150 is due in the next couple of years, the first time the bestselling pickup will be offered in a fully-electric form. Chevrolet, meanwhile, has an electric pickup in the works too, GM confirmed last week, tapping the automaker’s new Ultium platform.

Continue Reading

Cars

NHTSA: GM must recall 6m pickups and SUVs over Takata airbag danger

Published

on

GM will be forced to recall almost 6 million vehicles to repair potentially dangerous Takata airbags, after losing a years-long battle with the NHTSA to avoid the hugely expensive repairs. The automaker had argued that the recall – which covers some of its most popular SUVs and pickups – was unnecessary, given it had undertaken third-party tests to show that the airbag inflaters were not prone to dangerous or abnormal explosions.

The Takata airbag saga has become the most significant vehicle recall incident in the US, and forced the most manufacturer recalls. Commonly used across multiple brands, the inflators are designed to trigger in a crash and rapidly inflate the airbags themselves to support vehicle occupants.

However the chemicals inside the flawed inflators can degrade over time, particularly in conditions of high heat or high humidity. That in turn can cause an increase in force beyond the intended specifications, shattering the metal canister and releasing a spray of dangerous shrapnel as a result. There have been 27 deaths blamed on the inflators worldwide, 18 of which have been in the US, and hundreds of injuries.

GM’s argument was that the vehicles – based on the GMT900 platform from brands like Chevrolet, Cadillac, and GMC, and including the Avalanche, Escalade, Escalade ESV, Escalade EXT, Sierra 1500, Sierra 2500/3500, Silverado 1500, Silverado 2500/3500, Suburban, Tahoe, Yukon, and Yukon XL – actually used different inflator designs, integrated in different ways. It undertook third-party testing by Northrop Grumman’s OATK, among others, in the hope of demonstrating to the NHTSA that, unlike with other manufacturers, a full recall wasn’t necessary.

Now, after a four year back-and-forth between automaker and agency, the National Highway Traffic Safety Administration has denied GM’s request. “After reviewing GM’s consolidated petition, supporting materials, and public comments,” the agency said today, “NHTSA has concluded that GM has not met its burden of establishing that the defect is inconsequential to motor vehicle safety, and denies the petition.”

The decision will impact approximately 5.9 million vehicles, from model years 2007 through to 2014. Estimates peg the total cost to GM at $1.2 billion.

Despite GM’s validation of its changes to the Takata design and implementation, the NHTSA deemed the risk still too high. “Given the severity of the consequence of propellant degradation in these air bag inflators – the rupture of the inflator and metal shrapnel sprayed at vehicle occupants – a finding of inconsequentiality to safety demands extraordinarily robust and persuasive evidence,” Jeffrey M. Giuseppe, Associate Administrator for Enforcement at the agency, wrote. “What GM presents here, while valuable and informative in certain respects, suffers from far too many shortcomings, both when the evidence is assessed individually and in its totality, to demonstrate that the defect in GMT900 inflators is not important or can otherwise be ignored as a matter of safety.”

The automaker now has 30 days to submit a proposed schedule of how it plans to notify owners of the affected vehicles, and how it will launch and operate the recall process.

Continue Reading

Cars

2021 Ford Mustang Mach-E official EPA range confirmed

Published

on

Ford has final EPA range figures for its upcoming 2021 Mustang Mach-E, and there’s good news for those waiting for the imminent all-electric crossover. While the company had estimated range numbers for the new EV back when it unveiled it in late 2019, they’ve only been certified by the Environmental Protection Agency today. Turns out, Ford’s predictions were almost exactly on the dot.

The automaker had been targeting 230 miles for the Mustang Mach-E standard range RWD configuration, and 300 miles for the extended range RWD version. The EPA says that’s the case, as is it the 270 mile rating of the Mustang Mach-E extended range eAWD car.

The Mustang Mach-E standard range eAWD actually did ever so slightly better in its official rating. Ford had promised 210 miles; the EPA ranks it at 211 miles. Final testing for the Mustang Mach-E California Route 1 version of the electric crossover is still underway, with that configuration estimated at 300 miles.

It’s a note of good news in the final few weeks before Mustang Mach-E cars actually arrive with preorder customers. Ford says that customer deliveries should start in December 2020, though high-end versions of the EV – like the Mustang Mach-E GT – aren’t expected until 2021.

Though the range figures aren’t exactly the largest in the category, Ford’s argument has been that there’s more to driver satisfaction than just a big number. For a start, there’s ease of recharging. With up to 150 kW charging support (or 110 kW on the entry-level Select trim), assuming you can find a DC fast charger you should be able to add 52-61 miles of range in 10 minutes, depending on drivetrain configuration. Using the FordPass Charging Network, effectively an umbrella access several different third-party networks like Electrify America, actually finding those stations should be more straightforward too.

The Mustang Mach-E will be one of the few electric vehicles in the US to support Plug&Charge, too. That means, at a compatible charger such as those offered by Electrify America, drivers won’t even need to scan a card to begin the charging session. Instead, that digital handshaking – including authenticating the driver’s account – will all be done between the EV and the charger.

Ford’s other push has been around a more accurate range estimate for the dashboard. Range anxiety, after all, isn’t just about total miles of driving left, but uncertainty about whether the number displayed is actually accurate. Ford plans to not only use data from the individual EV itself, but crowdsource better estimates between cars.

The first iteration of Ford Intelligent Range will take into account things like past driver behavior and forecasted weather as it calculates how much driving you’ll be able to do before a recharge. Later, though, Ford plans to light up range data sharing, which will use the EV’s embedded modem to give anonymized feedback of how battery use was affected by things like speed, terrain, and climate conditions. That way, if your journey is going to take you on a new route where other Mustang Mach-E drivers have used more energy than might be expected for one reason or another, the car will proactively take that into account.

Continue Reading

Trending