Notice: This website is continuously updated as new conference proceedings and journal volumes are published. Consequently, the figures presented here may differ from those reported in our publications.
A brief overview of our methodology is provided on the Methods page, along with references to publications documenting our approach and links to the code and data used in each study.
The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates.
LSB: Local Self-Balancing MCMC in Discrete Spaces
Authors: Emanuele Sansone
ICML 2022 | Venue PDF | LLM Run Details | Input Tokens: 21,399 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 3,910 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on energy-based models and Markov networks show that LSB converges using a smaller number of queries to the oracle distribution compared to recent local MCMC samplers. We conduct experiments on energy-based models and Markov network, and show that LSB queries the oracle distribution in a more efficient way compared to recent local MCMC samplers. |
| Researcher Affiliation | Academia | Emanuele Sansone 1 1Department of Computer Science, KU Leuven, Belgium. Correspondence to: Emanuele Sansone <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 Local Self-Balancing (LSB). Algorithm 2 Fast Local Self-Balancing (FLSB). |
| Open Source Code | Yes | Code to replicate the experiments is available at https://github.com/emsansone/LSB.git. |
| Open Datasets | Yes | In particular, we train a RBM with 250 hidden units on the binary MNIST dataset using contrastive divergence and use this model as our base distribution for evaluating the samplers. |
| Dataset Splits | No | The paper mentions "burn-in iterations" and "iterations for sampling" but does not specify standard training/validation/test dataset splits with percentages or sample counts. |
| Hardware Specification | No | The paper describes the models and experiments performed but does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions the use of code and implies common machine learning frameworks, but it does not specify any software names with version numbers, such as Python versions or library versions (e.g., PyTorch 1.9). |
| Experiment Setup | Yes | Input: Learning rate γ = 1e-2, π = 1e-8, initial parameter θ0, burn-in iterations K and batch of samples N {ˆx(i)}N i=1 UX for k = 1 to K do ... Learning rate η = 1e-2 for SGD optimizer with momentum. Burn-in iterations K = 2000 (for Ising), K = 24000 (for RBM), K = 500 (for UAI). Iterations for sampling 30000 (for Ising), 120000 (for RBM), 10000 (for UAI). Batch size N = 30 (for Ising), N = 16 (for RBM), N = 5 (for UAI). MLP network with one hidden layer of 10 neurons (for ISING and RBM) and monotonic network with 20 blocks of 20 neurons (for UAI). |