Learning Physical Intuition of Block Towers by Example

Authors: Adam Lerer, Sam Gross, Rob Fergus

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental This data allows us to train large convolutional network models which can accurately predict the outcome, as well as estimating the block trajectories. ... Table 1 compares the accuracy for fall prediction of several deep networks and baselines described in Section 2.4. ... Fig. 6 compares Phys Net to 10 human subjects on the same set of synthetic and real test images.
Researcher Affiliation Industry Adam Lerer ALERER@FB.COM Facebook AI Research Sam Gross SGROSS@FB.COM Facebook AI Research Rob Fergus ROBFERGUS@FB.COM Facebook AI Research
Pseudocode No The paper describes network architectures and processes in prose and diagrams (e.g., Fig. 4), but does not contain pseudocode or explicitly labeled algorithm blocks.
Open Source Code Yes UETorch: We introduce an open-source combination of the Unreal game engine and the Torch deep learning environment, that is simple and efficient to use. ... The UETorch package can be downloaded freely at http: //github.com/facebook/UETorch.
Open Datasets No The paper describes generating a synthetic dataset: 'A simulation was developed in UETorch that generated vertical stacks of 2, 3, or 4 colored blocks in random configurations. ... A total of 180,000 simulations were performed...' However, it does not provide a direct link or citation for public access to this specific collected dataset.
Dataset Splits Yes A total of 180,000 simulations were performed, balanced across number of blocks and stable/unstable configurations. 12,288 examples were reserved for validation.
Hardware Specification No The paper does not specify the exact hardware components (e.g., specific CPU or GPU models, memory, or cloud instance types) used for running the experiments.
Software Dependencies No The paper mentions the use of 'Unreal Engine 4 (UE4)' and the 'Torch' framework, but it does not provide specific version numbers for these or any other software dependencies.
Experiment Setup Yes We replaced the final linear layer with a single logistic output and fine-tuned the entire network with SGD on the blocks dataset. Grid search was performed over learning rates.