Profile cover photo
Profile photo
Wayne Radinsky
18,528 followers -
Software Design Engineer
Software Design Engineer

18,528 followers
About
Posts

Post has attachment
"By combining state-of-the-art techniques in computer vision and reinforcement learning, our system enables simulated characters to learn a diverse repertoire of skills from video clips. Given a single monocular video of an actor performing some skill, such as a cartwheel or a backflip, our characters are able to learn policies that reproduce that skill in a physics simulation, without requiring any manual pose annotations."

"Our framework is structured as a pipeline, consisting of three stages: pose estimation, motion reconstruction, and motion imitation. The input video is first processed by the pose estimation stage, which predicts the pose of the actor in each frame. Next, the motion reconstruction stage consolidates the pose predictions into a reference motion and fixes artifacts that might have been introduced by the pose predictions. Finally, the reference motion is passed to the motion imitation stage, where a simulated character is trained to imitate the motion using reinforcement learning."
Add a comment...

Post has attachment
"The robot economy will run on blockchain." "Blockchain was first successfully implemented for crypto-currencies like bitcoin, creating mathematically protected trade operations independent of external administrators like banks or state bodies. Then, in 2015, the Ethereum platform was launched, allowing for smart contracts to be placed on the blockchain. These are contracts of arbitrary complexity that can be verified by a public network in the same way that cryptocurrency transactions are verified. They unite into one digital object the terms of a contract, and its execution."

"In our opinion, the robot economy should be built on these smart contracts. They naturally solve the issue of monitoring the fulfillment of obligations. They reduce friction among contracting parties. Information on transactions is verifiable and unchangeable. The unambiguous recording of information allows reliable reputation scores to be created. The blockchain can be organized so that network participants do not benefit from discrediting it."
Add a comment...

Post has attachment
DeepMind has open sourced a reinforcement learning toolkit called TRFL. "Today we are open sourcing a new library of useful building blocks for writing reinforcement learning (RL) agents in TensorFlow. Named TRFL (pronounced 'truffle'), it represents a collection of key algorithmic components that we have used internally for a large number of our most successful agents such as DQN, DDPG and the Importance Weighted Actor Learner Architecture."

"The TRFL library includes functions to implement both classical RL algorithms as well as more cutting-edge techniques. The loss functions and other operations provided here are implemented in pure TensorFlow. They are not complete algorithms, but implementations of RL-specific mathematical operations needed when building fully-functional RL agents."
Add a comment...

Post has attachment
"In this tutorial we are going to learn how to train deep neural networks, such as recurrent neural networks (RNNs), for addressing a natural language task known as emotion recognition." "We will cover the common best practices, functionalities, and steps you need to understand the basics of TensorFlow’s and PyTorch’s APIs to build powerful predictive models via the computation graph. In the process of building our models, we will compare PyTorch and TensorFlow to let the learner appreciate the strengths of each tool."

The author provided both TensorFlow and PyTorch notebooks.
Add a comment...

Post has attachment
"If delivering lunch wasn't enough, the robot will even keep the food hot or cold while it's on the way. Its called the Loomo delivery robot and the company has partnered with Meituan-Dianping, China's largest on-demand service provider in hopes to get their robots in more buildings across China."

"It has replaceable containers and can carry up to 110 pounds. Traveling up to 3 mph, the robot can work for about 8 hours before it autonomously returns to its charging station."

"When the robot arrives at an elevator, it will send instructions to the central control system of the elevator via a wireless network, and then the doors will open. Once the robot reaches it's intended destination, an office worker simply types their phone number into the touchscreen and out pops their lunch."
Add a comment...

Post has attachment
Roborace has pivoted from pure AI racing to human + AI teams. The first season will be in 2019.

"It was fascinating to see how quickly they could get the AI to adapt to the new information. Initially I had a lap time that was almost four seconds faster than my teammate's. Then they looked at the data from the runs and saw I was braking a lot later than the AI was, so [they] used that data. By day two, the AI was just a tenth slower."
Add a comment...

Post has attachment
"Researchers at the Université de Sherbrooke in Québec, Canada, have managed to equip a microelectromechanical system (MEMS) device with a form of artificial intelligence, marking the first time that any type of AI has been included in a MEMS device. The result is a kind of neuromorphic computing that operates like the human brain but in a microscale device. The combination makes it possible to process data on the device itself, thus improving the prospects for edge computing."

"The AI method the researchers demonstrated in their research, which is described in the Journal of Applied Physics, is something called 'reservoir computing.'" "Reservoir computing uses a dynamical system driven by the time-dependent input. The dynamical system is chosen to be relatively complex, so its response to the input can be fairly different from the input itself."
Add a comment...

Post has attachment
"One of the tools that have caught my attention this week is MedicalTorch (developed by Christian S. Perone), which is an open-source medical imaging analysis tool built on top of PyTorch. It contains a set of loaders, pre-processors and utility functions to efficiently and easily analyze medical images such as those acquired from magnetic resonance imaging (MRI) scans."

"In this post, I will summarize some of the functionalities offered by the medicaltorch library and how it can be used to conduct medical imaging analysis. Specifically, this will be a tutorial on how to perform spinal cord gray matter segmentation using a technique based on convolutional neural networks (CNNs)."
Add a comment...

Post has attachment
"Tesla CEO Elon Musk updated the timeline to release the company's new neural net computer, which they claimed will be the 'world's most advanced computer for autonomous driving'."

"They are now aiming for the new computer to be in production in about 6 months and it could result in a 500-2000% increase in operation per second, according to Musk."

"The release of this new computer with Tesla's own AI chip would be the culmination of a long project that Tesla started about 3 years ago as it anticipated a need for more computing power in its vehicles."
Add a comment...

Post has attachment
Tesla Autopilot V9 neural network improvements over V8. The 3 front cameras and 1 back camera feed in 2 frames of 1280x960 full color video into the neural network. The other side cameras are 640x416, one frame only, not full color. In the previous neural network, all the cameras were 640x416, 2 color channels, 1 frame at a time.

The same weight file is being used for all cameras. Previously V8 used separate weights for each camera. The neural network is an Inception V1 convolutional neural network, has 5x as many weights an V8, uses 18x as much processing power per camera, and processes 13x as much data.

Compared with Google's original V1 Inception network, Tesla's V9 camera network is 10x larger and requires 200x the computation power.

The network outputs a V360 object decoder, a back lane decoder, a side lane decoder, a path prediction pp decoder, and a 'super lane' decoder. These are passed to another subsystem.

"When you increase the number of parameters (weights) in an NN by a factor of 5 you don't just get 5 times the capacity and need 5 times as much training data. In terms of expressive capacity increase it's more akin to a number with 5 times as many digits. So if V8's expressive capacity was 10, V9's capacity is more like 100,000. It's a mind boggling expansion of raw capacity. And likewise the amount of training data doesn't go up by a mere 5x. It probably takes at least thousands and perhaps millions of times more data to fully utilize a network that has 5x as many parameters."

"This network is far larger than any vision NN I've seen publicly disclosed and I'm just reeling at the thought of how much data it must take to train it. I sat on this estimate for a long time because I thought that I must have made a mistake. But going over it again and again I find that it's not my calculations that were off, it's my expectations that were off."

"With these new changes the NN should be able to identify every object in every direction at distances up to hundreds of meters and also provide approximate instantaneous relative movement for all of those objects. If you consider the FOV overlap of the cameras, virtually all objects will be seen by at least two cameras. That provides the opportunity for downstream processing use multiple perspectives on an object to more precisely localize and identify it."

"I've been driving V9 AP2 for a few days now and I find the dynamics to be much improved over recent V8. Lateral control is tighter and it's been able to beat all the V8 failure scenarios I've collected over the last 6 months. Longitudinal control is much smoother, traffic handling is much more comfortable. V9's ability to prospectively do a visual evaluation on a target lane prior to making a change makes the auto lane change feature a lot more versatile. I suspect detection errors are way down compared to V8 but I also see that a few new failure scenarios have popped up (offramp / onramp speed control seem to have some bugs)."
Add a comment...
Wait while more posts are being loaded