Sequence Modeling with Multiresolution Convolutional Memory

Authors: Jiaxin Shi, Ke Alexander Wang, Emily Fox

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical evaluation covers sequential image classification and autoregressive generative modeling (CIFAR-10), reasoning on syntax trees (List Ops), and multi-label classification of electrocardiogram (PTB-XL).
Researcher Affiliation Academia Jiaxin Shi 1 Ke Alexander Wang 1 Emily B. Fox 1 2 1Stanford University 2CZ Biohub SF.
Pseudocode Yes We provide an example Py Torch implementation of a MULTIRESLAYER with the resolution fading Tree Select strategy (see Sec. 3.2) in Fig. 3.
Open Source Code Yes Py Torch code can be found in https://github.com/thjashin/multires-conv.
Open Datasets Yes We evaluate our model on the Sequential CIFAR-10 dataset, which has long been used as a standard benchmark for modeling long-range dependencies in RNNs. [...] long List Ops dataset from Tay et al. (2021). [...] PTB-XL (Wagner et al., 2020) is a publicly available dataset of electrocardiogram (ECG) time series.
Dataset Splits Yes We use the standard train and test split of the CIFAR-10 dataset and leave out 10% of the training set as the validation set.
Hardware Specification No The paper mentions 'hardware accelerators implementing convolutions' but does not specify any particular GPU models, CPU models, or other hardware details used for the experiments.
Software Dependencies No The paper refers to 'Py Torch code' but does not specify exact version numbers for PyTorch, Python, CUDA, or other relevant libraries.
Experiment Setup Yes We use the Adam optimizer with default hyperparameters and decoupled weighted decay (Loshchilov & Hutter, 2018). Dropout is applied after the GELU and gated linear activation functions whenever overfitting is observed.