The AI is designed to be non-invasive and uses a short language sample from a verbal cognitive test given to a patient. Using this sample, the AI model is able to predict the onset of Alzheimer’s with around 71 percent accuracy.

For comparison, standard clinical tests are correct approximately 59 percent of the time and take much longer to diagnose. Current tests analyse the descriptive abilities of people as they age for potential warning signs.

In a paper detailing IBM’s model, the company says it used data from the Framingham Heart Study.

The study first began in 1948 and spans the multiple generations required for building an AI to predict Alzheimer’s in healthy individuals with no other risk factors. 5,000 participants from Massachusetts and their families have been studied.

703 samples from 270 of the study’s participants were collected and analysed to create a dataset consisting of a single sample from 80 participants—half of whom developed Alzheimer’s symptoms before they reached 85.

The AI was trained on this dataset to spot Alzheimer’s signals such as the repetition of words and using short sentences with poor grammatical structures. IBM’s AI was able to correctly predict the onset of Alzheimer’s in every seven of ten cases.

IBM intends to expand the training of their model using more data to better reflect society including socioeconomic, racial, and geographic factors. The Alzheimer’s research is part of a broader IBM effort to better understand neurological health and chronic illnesses through biomarkers and signals in speech and language.

Around 5.5 million people in America alone are estimated to have Alzheimer’s, and some studies suggest it’s the third leading cause of death behind heart disease and cancer.

While there is no cure or prevention for Alzheimer’s yet, earlier diagnosis helps to prepare individuals and their families as much as possible. If treatments become available, Alzheimer’s will almost certainly be more effectively treated when caught earlier.

IBM published its research in The Lancet’s science journal EClinicalMedicine. Pfizer was disclosed as providing funding to obtain data from the Framingham Heart Study Consortium and supporting IBM Research’s involvement.

(Image: Jeff Rogers, global research lead for IBM Research’s Digital Health platform, at work in the IBM Home Health Lab.)

The post IBM’s latest AI predicts Alzheimer’s better than standard tests appeared first on AI News.

The AI-powered tool is able to analyse CT scans for signs of COVID-19 infection to help quickly diagnose the novel coronavirus which has caused havoc around the world.

With heroic medical staff under more pressure than ever caring for the huge influx of people suffering with COVID-19 – in addition to all the other ailments they have to treat – such an AI-powered tool could help to free up significant amounts of time.

M3 has been testing the solution in Japan since the end of March; with the aim of deploying it across hundreds of locations. 

Hospitals will send CT scans to M3’s system which will then return the results with a 1-5 scale indicating the likelihood of COVID-19 pneumonia.

Alibaba’s system has been used in Chinese hospitals – including in Wuhan, the expected source of the COVID-19 outbreak – for a while now. The Chinese tech giant claims its AI can diagnose COVID-19 within 20 seconds with an accuracy of 90 percent or higher.

On average, a doctor takes around 20 minutes to make a diagnosis once a CT scan is available. M3 has found that the system typically diagnoses in under a minute.

While finding the accuracy to be relatively high, M3 reports the accuracy falls short of the 90 percent claimed by Alibaba. Even at 90 percent, 100 patients in every 1000 risk being misdiagnosed.

However, reading COVID-19 scans is reportedly even tricky for skilled physicians – especially as the virus is still relatively new. An AI-powered system which frees up clinical time is sure to be welcomed by all hospitals.

Catching the smaller signs of COVID-19 early could even help with providing treatment to those who need it before they get seriously ill.

This isn’t the first time AI has been looked to for assistance in tackling the COVID-19 pandemic.

Earlier this week, researchers from WVU Medicine and the Rockefeller Neuroscience Institute said they were able to predict the onset of COVID-19 symptoms three days early using AI to analyse data from Oura’s wearable rings.

Back in April, researchers from Carnegie Mellon University launched an AI-powered voice analysis system which aims to determine whether someone is suffering from COVID-19 using just a website.

While it seems likely we’re going to be living with COVID-19 in our lives for the foreseeable future, AI technologies look ready to step in and help.

(Photo by Robina Weermeijer on Unsplash)

The post M3: Alibaba’s AI detects COVID-19 pneumonia in under a minute appeared first on AI News.

Loup Ventures asked each of the three main virtual assistants – Google Assistant, Alexa, and Siri – a total of 800 questions. The assistants understood almost every question, even if not all of the responses were correct/sufficient.

In terms of understanding the questions, these are the results:

• Google Assistant – 100 percent
• Alexa – 99.9 percent
• Siri – 99.8 percent

Loup Ventures’ say their question set it designed to comprehensively test a virtual assistant’s ability and utility. Questions are broken down into five categories:

1. Local – Where is the nearest coffee shop?
2. Commerce – Order me more paper towels.
3. Navigation – How do I get to Uptown on the bus?
4. Information – Who do the Twins play tonight?
5. Command – Remind me to call Jerome at 2 pm today.

This is the percentage of questions each assistant answered correctly:

• Google Assistant – 92.9 percent
• Siri – 83.1 percent
• Alexa –  79.8 percent

The results are a huge improvement over Assistant, Alexa, and Siri’s results last year.

In 2018, Loup Ventures found Google Assistant answered the most questions with an 86 percent success rate. This was followed by Siri at 79 percent, while Alexa trailed behind at just 61 percent.

Alexa’s jump in answering the question correctly from 61 percent last year to almost 80 percent this year is the most commendable performance improvement, even if Amazon’s assistant is still in last place overall.

The researchers explained that they’ve stopped including Cortana in their tests due to a strategy change from Microsoft earlier this year.

Microsoft said in January that it’s no longer attempting to compete with Alexa or Google Assistant in areas like smart speakers, but instead is repositioning Cortana more like a skill that can be embedded in services where she can be of assistance.

The post Google Assistant wins IQ test, but Alexa and Siri are catching up appeared first on AI News.

The company’s flagship P-series of smartphones focuses on camera innovations. With this latest device, we’re seeing that trend continue. There are clear enhancements to existing features in addition to a few new things for us to play with.

In the Mate 20, Huawei’s previous flagship, the company expanded the number of scenes and objects its AI could recognise to 5,000 different objects and 1,500 different scenarios across 25 categories.

While impressive, many ended up switching the feature off and Huawei even released an update to do so by default. Objects were often misrecognised and images ended up looking poorer.

Huawei has made clear improvements with the P30 Pro. Aside from one amusing occurrence where my dog was recognised as a panda – and I’m fairly sure he’s not – the AI recognition has excelled, and even surprised (in a good way.)

The standout moment was taking a shot of a flower. From a distance, the AI identified the shot was of some ‘greenery’ and made the colours punchier. Framing more of the flower caused the AI to switch over to ‘flowers’ and made it sharper with a nice depth-of-field bringing the object into focus while blurring the backdrop.

Moving closer still, to capture small details of the flower, caused the AI to switch to ‘super-macro’. This mode uses the P30 Pro’s incredible macro capabilities of taking images just 2.5cm away.

Low-light photography is another area where Huawei is harnessing AI to elevate its performance. The AI image stabilisation allows the user to capture long-exposure photographs without the need for a tripod.

In the image below, both shots were taken from the same place in near pitch-black conditions:

One implementation of AI we’re unconvinced about is the smartphone’s use for HDR. As you can see below, the feature led to a washed-out image which you’d not expect from a HDR shot:

Huawei has issued updates in the past to improve the camera performance of its devices, so we’ll be keeping an eye out for one to address this. For now, it seems better to keep the HDR feature switched off.

Ahead of the P30 announcement, Huawei hyped the device’s zooming capabilities. Based on our tests, we can understand why:

In the first image, I’ve had to mark where the small daffodil is that we’re zooming in on. At 5x, due to being an optical zoom, we’re seeing no quality loss. At 10x, Huawei makes use of AI for a hybrid zoom that it claims is ‘lossless’ (we think there’s some loss, but it’s negligible.)

At 50x zoom, we’re seeing a noticeable loss of image quality. However, that is an incredible zoom for a smartphone camera and is only enabled through AI stabilisation. The resulting image is still recognisable even though it won’t be winning awards. Considering how difficult it is to pick out the flower by eye in the non-zoomed image, it’s a commendable performance.

Moving slightly away from photography into videography, the P30 Pro uses AI in combination with OIS (Optical Image Stabilisation) to keep things steady in rough conditions. We put it through some fairly extreme tests of waving the device while recording, and even going for a jog.

In both scenarios, the AI stabilisation performed admirably. While it was not perfectly stable, the result was viewable and it’s hard to imagine stabilisation getting much better (although, I have been surprised in the past.)

A clip of the AI video stabilisation is available in our YouTube review below:

Finally, we’ve benchmarked the Huawei P30 Pro against the company’s previous flagship – the Mate 20 Pro. Here are the results:

Your eyes aren’t deceiving you, the P30 Pro had a slightly lower overall result. It’s a surprise given both use the same NPU, CPU, and GPU. In fact, the P30 Pro has 2GB more RAM to play with over the Mate 20 Pro (8GB vs 6GB).

Benchmarks only tell a small part of the story and this one is overwhelmingly positive. Huawei has excelled itself and set the bar for AI on a smartphone with the P30 Pro.

deepgeniusai.com/">AI & Big Data Expo events with upcoming shows in Silicon Valley, London, and Amsterdam to learn more. Co-located with the IoT Tech Expo, , & .

The post Huawei P30 Pro: AI turns anyone into a skilled photographer appeared first on AI News.

If you’ve ever read a comprehensive smartphone review, you’ve likely heard of AnTuTu. The company’s smartphone benchmarking tool is often used for testing and comparing the CPU and 3D performance of devices.

With dedicated AI chips now appearing in devices from the mid-range to flagships, AnTuTu has decided it’s time for a benchmark to determine their performance.

In a blog post, AnTuTu says its benchmark uses two categories – ‘Image Classification’, and ‘Object Recognition’.

AI News tested AnTuTu’s benchmark on a Huawei Mate 20 Pro which currently ranks second on AnTuTu’s general performance leaderboard for Android devices. Huawei often brags about the AI performance of its flagship devices.

The first test classifies 200 images as fast as possible using the Inception v3 neural network:

In the second test, 600-frame video is reviewed using the MobileNet SSD neural network:

AnTuTu then delivers an overall benchmark score, along with the scores for each category.

Here is how our Mate 20 Pro fared:

• Overall – 65,222
• Image Classification – 41,717
• Object Detection – 23,505

Each of the categories is further broken down into scores for ‘speed’ and ‘accuracy’. If accuracy is traded for speed, then a lower score will be given.

AnTuTu says this helps to prevent cheating by devices processing the data fast but without providing the right answers. Smartphone manufacturers have been caught artificially-inflating their benchmarks in the past; so it provides added confidence in the results.

For a general look at the AI features in the Mate 20 Pro, see our video below:

You can download the AI benchmark from AnTuTu here.

deepgeniusai.com/">AI & Big Data Expo events with upcoming shows in Silicon Valley, London, and Amsterdam to learn more. Co-located with the IoT Tech Expo, , & .

The post AnTuTu’s latest benchmark tests AI chip performance appeared first on AI News.