Learning Physics from Dynamical Scenes

From Ref:ullman2014learning:

Humans acquire their most basic physical concepts early in development, but continue to enrich and expand their intuitive physics throughout life as they are exposed to more and varied dynamical environments. We introduce a hierarchical Bayesian framework to explain how people can learn physical theories across multiple timescales and levels of abstraction.

The easy way

To run the computational experiments from the paper, install Docker, then:

mkdir -p inference/world1
docker run -it -v $(pwd)/inference:/data/ullman-physics2015/inference/ stuhlmueller/ullman-physics 1 2

Running this writes the inference results for world 1, scenario 2 to the inference/world1 directory. It won’t print out anything while it is running, and this will take a long time, but you can check intermediate results, e.g. by running tail -f inference/world1/world1_2 in the same directory (in a different terminal).

You can also create a shell in an environment that has everything needed to run the model:

1. docker run -it --entrypoint bash stuhlmueller/ullman-physics
2. e.g. ikarus --script run-inference.ss 1 2 (as before, this will only print to a file; if we don’t give the -v argument as above, this file will only be in the Docker container and not on your computer)

The hard way

The implementation of this computational model requires Ikarus scheme. Follow these instructions to install Ikarus and run the model:

1. Log into a compatible computer. We recommend launching a 32-bit Ubuntu version such as ubuntu-trusty-14.04-i386-server-20140724 on EC2. To find this machine image, search for ami-07b4b742 under Community AMIs. After launch, log in:

    ssh -i my-keypair.pem ubuntu@ec2-...compute.amazonaws.com
   

2. Install required packages:

    sudo apt-get update
    sudo apt-get install build-essential autoconf m4 automake perl
    sudo apt-get install libgsl0-dev libffi-dev libgmp3-dev
    sudo apt-get install git unzip bzr
   

3. Install Ikarus Scheme:

    wget http://stuhlmueller.org/data/ikarus.zip
    unzip ikarus.zip
    cd ikarus.dev-derick
    ./configure --enable-libffi
    make
    sudo make install
    cd ..
   

4. Install scheme-tools:

    git clone https://github.com/stuhlmueller/scheme-tools.git
    export IKARUS_LIBRARY_PATH=~/scheme-tools/:$IKARUS_LIBRARY_PATH
   

5. Download the “Learning Physics” Scheme code:

    wget http://stuhlmueller.org/data/ullman-physics2015.zip
    unzip ullman-physics2015.zip
    export IKARUS_LIBRARY_PATH=~/ullman-physics2015/:$IKARUS_LIBRARY_PATH
   

6. Run the Scheme code (in this example: world 1, scenario 2):

    cd ullman-physics2015
    mkdir inference
    mkdir inference/world1/
    ikarus --script run-inference.ss 1 2
   

The inference results are written to inference/world1/world1_1.

A note on runtime: A forward-step of simulating physics is on the order of a second. However, depending on your setup and the number of theories being inferred, the inference step may take several hours.

References:

• Cite:ullman2014learning