End-to-end Differentiable Clustering with Associative Memories
Authors: Bishwajit Saha, Dmitry Krotov, Mohammed J Zaki, Parikshit Ram
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our evaluations on varied datasets demonstrate that Cl AM benefits from the selfsupervision, and significantly improves upon both the traditional Lloyd s k-means algorithm, and more recent continuous clustering relaxations (by upto 60% in terms of the Silhouette Coefficient). |
| Researcher Affiliation | Collaboration | 1CS Department, Rensselaer Polytechnic Institute, Troy, NY, USA 2MIT-IBM Watson AI Lab, IBM Research, Cambridge, MA, USA 3IBM Research, Yorktown Heights, NY, USA. |
| Pseudocode | Yes | Algorithm 1: Cl AM: Learning k prototypes for clustering a d-dimensional dataset S Rd into k clusters with T time-steps for N epochs, with inverse temperature β, learning rate ϵ, time step dt and time constant τ. |
| Open Source Code | Yes | The code is available at https://github.com/bsaha205/clam. |
| Open Datasets | Yes | To evaluate Cl AM, we conducted our experiments on ten standard benchmark data sets. The datasets are taken from various sources such as Yale from ASU feature selection repository3 (Li et al., 2017), USPS from Kaggle4 (Hull, 1994), Fashion-MNIST from Zalando5 (Xiao et al., 2017), GCM from Chakraborty et al. (2021) and the rest of the datasets from the UCI machine learning repository6 (Dua et al., 2017). |
| Dataset Splits | No | The paper mentions hyperparameter tuning and learning rate reduction based on training loss, implying a validation process (e.g., "Reduce LR patience (epochs) 5"), but it does not explicitly provide specific details about validation dataset splits (e.g., percentages or sample counts). |
| Hardware Specification | Yes | We train Cl AM on a single node with 1 NVIDIA Geforce RTX 3090 (24GB RAM), and 8-core 3.5GHz Intel Core-i9 CPUs (32GB RAM). |
| Software Dependencies | No | The paper mentions using "Tensorflow (Abadi et al., 2016)" and "scikit-learn (Pedregosa et al., 2011)" but does not provide specific version numbers for these software dependencies. |
| Experiment Setup | Yes | Hyperparameters are tuned for each dataset to find the best result. Full details of the hyperparameters used in our model are given in Table 5. ... Table 6 shows the best hyperparameters values for different datasets used in Cl AM. ... Table 7 provides a brief description of the hyperparameters and their roles in the baseline schemes. |