The lights dim and the music swells as an elite competitor in a silk robe passes through a cheering crowd to take the ring. It’s a blueprint familiar to boxing, only this pugilist won’t be throwing punches.
Each match of the International consists of two 5-player teams competing against each other for 35-45 minutes. In layman’s terms, it is an online version of capture the flag. While the premise may sound simple, it is actually one of the most complicated and detailed competitive games out there. The top teams are required to practice together daily, but this level of play is nothing new to them. To reach a professional level, individual players would practice obscenely late, go to sleep, and then repeat the process. For years. So how long did the AI bot have to prepare for this competition compared to these seasoned pros? A couple of months.
[Richard]’s project is based on the EOgma Neo machine learning library. Using a type of machine learning known as Sparse Predictive Hierarchies, or SPH, the algorithm is first trained with user input. [Richard] trained the model by driving it around a small track. The algorithm takes into account the steering and throttle inputs from the human driver and also monitors the feed from the Raspberry Pi camera. After training the model for a few laps, the car is then ready to drive itself.
Fundamentally, this is working on a much simpler level than a full-sized self-driving car. As the video indicates, the steering angle is predicted based on the grayscale pixel data from the camera feed. The track is very simple and the contrast of the walls to the driving surface makes it easier for the machine learning algorithm to figure out where it should be going. Watching the video feed reminds us of simple line-following robots of years past; this project achieves a similar effect in a completely different way. As it stands, it’s a great learning project on how to work with machine learning systems.
[Roland]’s workflow consists of breaking up a recording from his backyard into one second clips, loading them in to a Python program and running some machine learning code to determine whether the clip is a recording of a bat or not and using this to determine the number of bats flying around. He uses several Python libraries to do this including Tensorflow and LibROSA.
The Python code breaks each one second clip into twenty-two parts. For each part, he determines the max, min, mean, standard deviation, and max-min of the sample – if multiple parts of the signal have certain features (such as a high standard deviation), then the software has detected a bat call. Armed with this, [Roland] turned his head to the machine learning so that he could offload the work of detecting the bats. Again, he turned to Python and the Keras library.
Machine is an IDE for building machine learning systems using TensorFlow. You can sign up for the alpha, but first, have a look at the video below to see what it is all about.
You’ll see in the video, that you can import data for a model and then do training (in this case, to find a mustache in an image). You’ll see the IDE invites an iterative approach to development since you can alter parameters, run experiments, and see the results.
The IDE syncs with “the cloud” so you can work on it from multiple computers and roll back to previous results easily. We don’t know when the IDE will leave alpha status (or beta, for that matter), but the team’s goal is to release a free version of Machine to encourage widespread adoption.
The machine learning algorithm needed to be trained to identify the relevant parts of surgical videos. To do this, the laparoscopic surgeries being investigated were split up into distinct stages, each relating to a different part of the surgical process. Researchers would then watch recordings of prior surgeries and mark the start of each stage. This data was used to train the model which was then used to sift through other recordings to capture the key moments of each surgery.
The time-saving advantages of such technology could be applied to a great many fields – such an algorithm could be put to great use to sort through hours of uneventful security footage looking for anomalies, or rapidly cut together holiday footage so you only have to see the good parts. We’d love to see the researchers release footage showing the algorithm’s work – thus far, all we have to go off is the project paper.
Neural networks are all the rage right now with increasing numbers of hackers, students, researchers, and businesses getting involved. The last resurgence was in the 80s and 90s, when there was little or no World Wide Web and few neural network tools. The current resurgence started around 2006. From a hacker’s perspective, what tools and other resources were available back then, what’s available now, and what should we expect for the future? For myself, a GPU on the Raspberry Pi would be nice.
I had great fun writing neural network software in the 90s, and I have been anxious to try creating some using TensorFlow.
Google’s machine intelligence framework is the new hotness right now. And when TensorFlow became installable on the Raspberry Pi, working with it became very easy to do. In a short time I made a neural network that counts in binary. So I thought I’d pass on what I’ve learned so far. Hopefully this makes it easier for anyone else who wants to try it, or for anyone who just wants some insight into neural networks.