Awesome World Model Hub Papers · Datasets · Projects
← Back to papers

Do Transformer World Models Give Better Policy Gradients?

ICML 24 2024 51.5 method

TLDR

Transformer world models create circuitous gradient paths; proposed Actions World Models (AWMs) improve policy gradient optimization for long-horizon tasks.

Reasoning

The paper identifies a specific limitation of transformer world models (circuitous gradients) and proposes a novel architecture (AWMs) to address it, with empirical results on realistic tasks. However, the abstract lacks details on methodology, limitations, and explicit real-world validation, making it hard to assess generalizability.

Read-first score

Read-first score 51.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 34.

Recency 8%
75.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2024

Methodology quality 25%
70

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,evaluation,result

Topical relevance 42%
48.6

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 25%
30

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

Methodology anchor

Rank sensitivity

Stability: volatile; rank range: 275.

Keyword Scores

world model
10
model-based reinforcement learning world model
8
world dynamics prediction
6
world simulator
5
generative world model
2
interactive world model
2
video world model
1

Deep Analysis

Innovations

  • Identification that commonly-used transformer world models produce circuitous gradient paths detrimental to long-range policy gradients
  • Proposal of Actions World Models (AWMs) designed to provide more direct routes for gradient propagation
  • Integration of AWMs into a policy gradient framework that underscores the relationship between network architectures and policy gradient updates
  • Demonstration that AWMs can generate optimization landscapes easier to navigate than those from the simulator itself

Methodology

The authors propose Actions World Models (AWMs), a class of transformer-based world models designed to provide direct gradient propagation paths. They integrate AWMs into a policy gradient framework that explicitly relates network architecture to policy gradient updates. Evaluation is conducted on realistic long-horizon reinforcement learning tasks, comparing against competitive baselines.

Key Results

AWMs generate optimization landscapes that are easier to navigate even when compared to those from the simulator itself, allowing transformer AWMs to produce better policies than competitive baselines in realistic long-horizon tasks.

Tags