PATENT DROP: DeepMind’s code creator
Discover more from PATENT DROP
PATENT DROP: DeepMind’s code creator
Plus: Meta fixes VR rendering; Boeing finds the faults
Happy Thursday and welcome to Patent Drop!
Today, we’re digging into a patent from Google DeepMind for neural networks that can create computer code; a filing from Meta to make the metaverse easier to see; and Boeing’s tech to predict when its planes need a check-up.
Let’s get into it.
DeepMind gives developers a lift
Google DeepMind may be adding to the company’s AI-coding patent portfolio.
Google’s AI division filed a patent application for a system that generates computer code from “task descriptions” using neural networks. In simple terms, the system allows users to write out exactly what task they need done, and a generative neural network spits out one or more computer programs that can execute that task.
To break it down: A user sends task description inputs into this system, which generates several “candidate computer programs” to perform the task. The system will run those computer programs and, based on whether or not they can perform the task, will synthesize multiple candidate programs to perform it.
Code that solves a specific task can be difficult to create, as “single character edits can completely change program behavior,” DeepMind noted. “Even if they don’t cause crashes, solutions can look dramatically different even for the same problem.”
But DeepMind’s neural networks essentially apply a fine-tooth comb to this process, breaking down the problem a user wants to solve and quickly recalling lines of code from a “huge structured space” (a.k.a. a database of coding sequences). This can help a developer rapidly write highly specific code with less headaches and eye strain trying to pick out their tiny mistakes.
The system also can train the neural network on code sequences, then “fine-tune” that training on data that’s specific to the task, thereby making it able to create accurate code even if the amount of hyper-specific data is limited.
A filing like this from Google’s AI subsidiary (now AI division) isn’t a huge surprise. Google has touted AI integrations throughout Google Workspace and plans to supercharge its search engine with AI. The company also offers coding assistance with Duet AI for Google Cloud.
DeepMind, meanwhile, debuted AlphaCode last year, an AI-assisted tool for competitive programming, which isn’t available to the public yet. While competitive coding tactics aren’t often used in real-world programming, Alpha Code represents a “very impressive first step,” as it was able to beat out 54% of participants in programming competitions, said Rijul Gupta, co-founder of AI communications company Deep Media.
“When I read this patent, a lot of the stuff it talks about specifically makes me think of AlphaCode,” Gupta said. “What this shows is that AlphaCode and DeepMind in general are moving towards this system of a general AI with specific applications, and one of those would be a code assistant tool.”
This patent in particular, Gupta said, seems to represent a step toward a consumer or enterprise-grade product that could accelerate programmer capabilities by “as much as 10 to 100 times their normal amount.”
Meanwhile, Google has filed for several other patents in this vein: The company has sought to patent a system that automates parts of the computer programming process, a machine learning model that can create a user interface, and an AI-based tool that can create and publish a “viable running app.”
Taken together, these filings could point to a fully loaded, AI-enabled development toolkit that serves engineers of all skill levels, from the no- and low-code hobbyist developers to the full-time engineers who are looking for a way to get their job done quicker.
But Google could have some fierce competition in this area, with its biggest competition coming from Microsoft and OpenAI. Microsoft-owned GitHub operates an AI coding assistant called GitHub Copilot. OpenAI launched Code Interpreter in July, and GPT-4 itself is capable of writing solid code. Plus, Gupta noted, startups are likely working on similar coding assistants.
“The challenge is going to be a generalized solution,” said Gupta. “I think AlphaCode, Copilot and ChatGPT are working towards a generalized code assistant solution.”
–Nat Rubio-Licht
Meta’s quest to make VR crisper
Meta wants to make virtual scenes even more realistic.
The social media behemoth filed a patent application for methods that would reduce visual glitches when using virtual reality headsets.
The filing shows that Meta is keen to patent several techniques that make rendering its VR content smoother while using foveated rendering – a technique that uses eye tracking within a headset in order to display higher-quality graphics where the wearer is looking, and lower-density graphics on the periphery.
Meta’s patent is for technical tweaks that would likely go unnoticed by the user, but are designed to make rendering in its VR headsets better and smoother.
Here’s how the tech works: Meta’s custom GPU creates the image in the headset, then divides it into different sections, or tiles. These tiles are key to how Meta’s VR computers will decide what to render – the patent notes the headset will identify the tiles where the person’s gaze is, as well as some in the surrounding area, by using eye tracking.
Then, Meta’s headset computer will tell the GPU which tiles in the center of the person’s view to render in high detail (or, as Meta puts it, “high pixel-density”). Surrounding tiles will be rendered in “low-pixel density,” while those in between are produced with a mix of detail.
Most interestingly, Meta’s patent also said its headset computer can render border tiles in both high and low-pixel density, effectively duplicating them. This allows the headset to “blend” the border tiles with low and high density to “create a smooth transition.” This not only adds to the idea of realism in VR, making it feel more natural, but also saves Meta valuable energy.
Foveated rendering is a clutch way for companies like Meta to save precious computational power. VR is not yet energy-efficient, and some AI chip processors like those made by Intel claim to use 1,000 times more computing power than the average CPU. And the demand for computational energy when it comes to using VR is only expected to rise as the graphics get better. Remember how long it took for Meta founder Mark Zuckerberg to gleefully announce his Horizon Worlds metaverse avatars finally had legs? Now imagine the technical lift it’d take to transform that rough, animated physique into a realistic humanoid.
While Meta’s recently been in the headlines for its recent work on Llama 2, its newest AI large language model built to rival ChatGPT, the company’s not forgotten Zuckerberg’s ambitions to become the destination for virtual reality: “Management remains focused on bringing metaverse to the mainstream,” said Derek Higa, director and research analyst at William O’Neil.
Since the beginning of 2022, Meta’s Reality Labs has lost $21 billion, which management said it expects to increase. It’s got a stark sheet of competitors to face down, including Valve, HTC, Sony and Google. “I think success will reside in user adoption,” Higa noted, adding that Meta sold around 9 million Quest headsets in 2022, a roughly 40% annual growth. Still, Higa said “there really isn’t a ‘killer app’ for VR/AR yet which will really drive user growth to the next level.”
A choppy metaverse, lackluster graphics and dearth of community have left the virtual reality market stagnated. Last year, U.S. sales of VR headsets dropped 2% to $1.1 billion, while worldwide shipments sank by 12%, according to NPD Group.
Besides pricing, could graphics processing really make the difference between a buyer choosing a Meta headset over the others? That remains to be seen, but with some analysts speculating that motion sickness from choppy rendering is one factor hindering adoption of VR, smoothing out the experience could be Meta’s best bet.
– Samson Amore
Boeing predicts wear and tear in the air
Boeing wants to know if its planes are getting a little too rickety.
The aerospace manufacturer is seeking to patent a system for “component fault prediction.” Essentially, this system uses a prediction model to detect the likelihood that a vehicle component needs to be replaced, providing a recommendation for “addressing the predicted removal” of bad components.
This system collects tons of sensor data created by an aircraft during flight, including speed, temperature, and altitude. The model then weeds out data that is highly correlated — the things that are obvious — such as temperature and humidity, combining those pairs into individual data points to feed the system only non-correlated data.
What’s particularly interesting is the way the system arrives at its answers. Boeing’s system reduces the amount of data needed to make accurate predictions, allowing it to rapidly make decisions with reduced computational power. This is because, for conventional analysis, the number of data points to monitor can “easily reach into the hundreds of thousands,” Boeing noted.
Once the model decides that a part needs to be replaced, the patent notes a couple of options to amend the situation. For one, it could send that prediction to a “maintenance/parts inventory system,” which in turn instructs autonomous robots or systems to do the replacement work, reducing the time and manpower needed for upkeep.
If this prediction happens mid-flight, the system may reroute the aircraft to the closest airport that has that component.
Though the patent focuses specifically on component replacement, the prediction model itself could actually be applied in a number of different scenarios, said Rhonda Dibachi, CEO of Manufacturing-as-a-Service company HeyScottie. The model, she said, is basically just a way to reduce the amount of data needed to get to the same answer.
“The patent itself is for the algorithm for reducing a huge dataset into an actionable one,” Dibachi said. “I’m surprised it’s as narrow as it is in its application, because this algorithm can be applied to anything where you have a huge, unruly set of data.”
While the scope of Boeing’s application is quite narrow, the aerospace manufacturer may not be the only person to have figured out this method, Dibachi noted. Several academic papers have been written on the topic, specifically seeking to efficiently predict breakdowns using non-correlated data. Boeing’s confined approach may make it more likely to actually secure the patent.
Whether or not Boeing secures this patent, this tech could stand to save the company tons of time and resources, Dibachi said. The patent states that this tech has the potential to arrive at its predictions 10 times faster than any other kind of sorting algorithm, potentially allowing for corrective action to be taken preventatively, not reactively. Plus, this tech doesn’t require industry knowledge or a specialist to interpret the data, she said.
The cons are similar to that of any AI model, Dibachi said: AI is data-driven, and if the data spells out something that isn’t necessarily true, it can lead to false positives and false negatives, she said.
“You don’t know if it’s making a totally random prediction on something where the cards just happen to fall correctly,” Dibachi said. “Correlation is not causation, right?”
And given that Boeing is one of the biggest providers of commercial jets in the world, implementing this tech comes with an added layer of liability, and additional responsibility in getting it right — or at least having proper fail-safes in place.
—Nat Rubio-Licht
Extra Drops
Some other fun patents we wanted to share.
Walmart wants to know what hurts. The retailer is working on a conversational AI-based “symptom checker,” potentially adding to its pharmaceutical offerings.
Tesla wants to know you’re clicked in. The auto manufacturer wants to patent a system for “improper seat belt usage detection,” which sounds the alarm if you’re not buckled in correctly.
Nvidia wants to know if you have an unfair advantage in esports. The company wants to patent a system for game cheating detection based on “mouse lifting,” tracking if mouse movement data is “non-human.”
What else is new?
An AI model can accurately predict a VR user’s height, age, marital status and more based on the data these headsets collect, researchers found.
Alibaba reported solid earnings, with revenue growing by 14% and net income growing 51% in the recent quarter. The uptick is the biggest the company’s seen since September 2021.
You will soon be able to call people on X, formerly Twitter, CEO Linda Yaccarino confirmed, claiming that adding a video chat feature is part of its goal to become an “everything app.”
Have any comments, tips or suggestions? Drop us a line! Email at admin@patentdrop.xyz or shoot us a DM on Twitter @patentdrop.
 | A guest post by
|
 | A guest post by
|