Neuralink Update, Apple Self-Driving Car Patent, ProteinGANs : AI-CV News #1

In todays news, we are going to be talking about the ,insane update from Neuralink – With a ,monkey that can play pong with his mind, that’s right a monkey, Apples ,3D Voxel for Autonomous Driving, State of the art Multi-object tracking, Protein GANs, and a computer vision bot for ,grocery inventory management. So lets get into it. So AI-CV news is a brand new segment here on Augmented Startups to keep you guys updated with the ,latest in AI and Computer Vision. I’d really appreciate if you guys could comment down below if you’d like more of this. Also ,smash that like button and sub, it will really help me to beat the YouTube Algorithm and get me to 100’000 Subs. Lastly all the sources and mentioned in this video and AI courses will be in the links down below

Now onto the news.

State of the Art Multi-Object Tracking

GSDT: GNNs for Simultaneous Detection and Tracking - (analyticsindiamag.com)

So we all heard about CNN’s, no…not the news broadcaster, but rather convolutional neural networks. Now what the hell are GNN’s. Well they are called Graph Neural Networks and now they have been used in a model called GSDT which stands for GNNs for Simultaneous Detection and Tracking.

So essentially the authors from the Robotics Institute of Carnegie Mellon University have developed an end-to-end trainable Multi-object tracking architecture, which is said to achieve ..state of the art results in various public multi-object datasets like MOT15,16,17 and 20.

Whats really cool is that it can be trained and optimized as a whole and you can use GSDT to detect multiple objects and associate them simultaneously.

Trax

Trax Raises $640M for its Computer Vision-based Grocery Inventory Management (thespoon.tech)

If you ever worked in retail, you must know that inventory resupply and management can be quite tricky. I mean as a customer its really frustrating when there is something that you want and you drove all the way out of your comfy home to buy this one item that you want but you find out that it is either out of stock or missing. Very annoying!

A company called Trax has raised around $640 Million for its computer vision based Grocery inventory management system.

 Now this Trax system uses a combination of camera installations, shelf-scanning robots smooshed all together with computer vision to monitor products on store shelves. Product Images are then sent to Trax’s cloud-based machine learning system to analyze and identify when inventory is running low on shelves, or otherwise when items are misplaced. 

But this tech is not only good for just management of inventory. I can imagine that it can be used with some even more Intelligence to then do some inventory forecasting, so you know when to order goods, like toilet paper during COVID.

Apple VoxelNet Patent

Apple has won a Project Titan patent relating to techniques for implementing a 3D voxel feature detection network - Patently Apple

Accurately detecting 3D objects in point clouds may be considered a central problem in many real-world applications, such as autonomous driving, navigation, as well as augmented and virtual reality.

Apple have just recently secured a patent on Voxel Net which is said to be superior to LiDAR for autonomous driving-WOW. If you aren’t already aware, Apple has been working on Project titan which is their From the ground up, Self Driving Car. the iCar, Apple Car?...

Anyways this invention patent describes an approach for implementing a 3D voxel feature learning/detection network that may be configured to learn voxel features automatically from raw point cloud data, like a LiDAR point clouds. So this specific feature network allows the capture of subtle 3D shape information that is typically lost due to quantization or other processing steps in older systems.

A 3D voxel feature learning/detection network may improve prediction stability over time and may also improve mean average object detection precision also abbreviated as mAP.

They have conducted a bunch experiments on the KITTI car detection tests that show that the system may outperform state-of-the-art LiDAR based 3D detection methods by a large margin and may even surpass the accuracy of approaches fusing both RGB images and LiDAR. On more challenging tasks, such as 3D detection of pedestrians and cyclists, the system also demonstrates good performance due to its capability in effectively capturing various geometries. The future is now, or at least whenever Apple decides to launch their autonomous vehicle,… I wonder if we could jailbreak it.

ProteinGAN:

ProteinGAN: A generative adversarial network that generates functional protein sequences (phys.org)

Researchers at Biomatter Designs, and Chalmers University of Technology have recently developed ProteinGAN, which is a generative adversarial network otherwise known as GAN that can process and learn' different natural protein sequences. What this means is that this network uses the information it acquired to generate new functional protein sequences.

Now why is this significant? Well, living in a Pandemic, you can obviously see why. They tweeted that the AI generated proteins will help speed up drug discovery and has a lot of potential.

I mean protein generation is not easy. If one wants to make proteins aligned with human needs, one needs to correctly understand the order of amino acids and the given astronomical number of possibilities in making these proteins – Stated by Aleksej Zelezniak. Hope I pronounced that correctly.

Proteins are commonly used in our daily lives and are included in countless products, from washing powders to therapies against cancer and coronavirus. They are made of 20 amino acids that are arranged in different sequences and their order determines a protein's function."

Creating functional protein sequences is a very daunting task, as even a slight alteration in a given sequence can render a protein non-functional which can result in harmful and undesirable effects like cancers and other diseases.

So lets hope they can get it right and hopefully… develop a safe drug to cure COVID-19. I won’t lie, iv been itching to right travel now.

Neuralink monkey – smartphone control.

And then finally Elon musk tweeted with a video of a monkey that can ..wait for it…play pong with his mind. ΔWow this is [laughter]insane. So with a Neuralink chip implanted in the monkeys brain, the chip was able to decipher the commands that the brain was sending to the hand which was obviously controlling the joystick to the game. And now here’s the cool part, after calibration, they disconnected the controller but the monkey was still able to play pong by just using his mind. The inputs were all from the Neuralink interface! []OMG!!

Elon also tweeted that the product will enable someone with []paralysis to use a smartphone with their mind, faster than someone using thumbsΔ and eventually to walk again. Like the guy from 2 minutes papers YT Channel would say. “What a time to be alive!”

Augmented Startups

So if you enjoyed this segment and would like your weekly news update in the world of AI and Computer Vision, let me know in the comments down below. Also a like and sub to the channel will really help me beat the YouTube algorithm and get me to 100K subs much sooner, it will be much appreciated 😊.

If you are interested in enrolling in my 6-in-1 AI Mega course, where I teach Yolov4, Detectron2, Object Segmentation and Tracking, OpenCV AI Kit, AI on Raspberry Pi, then you can enroll at the link down below.