Reusable Slotwise Mechanisms

Authors: Trang Nguyen, Amin Mansouri, Kanika Madan, Khuong Duy Nguyen, Kartik Ahuja, Dianbo Liu, Yoshua Bengio

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate the superior performance of RSM compared to state-of-the-art methods across various future prediction and related downstream tasks
Researcher Affiliation Collaboration Trang Nguyen Mila Quebec AI Institute FPT Software AI Center Montreal, Canada
Pseudocode Yes Algorithm 1 Reusable Slotwise Mechanisms
Open Source Code Yes Video visualizations of our experiments are provided in our repository github.com/trangnnp/RSM
Open Datasets Yes OBJ3D [29] contains dynamic scenes of a sphere colliding with static objects.
Dataset Splits Yes Regarding the rollout frames K and the video length V in Table 5, we predict and consider K future frames from the last burn-in steps for training, whereas, we produce the total of V frames in the inference time, including T burn-in and V T rollout steps.
Hardware Specification Yes Each experiment is trained on 4 V100-GPUs with 12 CPUs, using a distributed data-parallel training strategy.
Software Dependencies No The paper mentions training on GPUs and CPUs but does not specify software dependencies with version numbers.
Experiment Setup Yes Table 5: Summary of experiments configuration, including the configuration of datasets, training process, and RSM.