Research map / Reinforcement learning / Visual and representation RL
LECO: Learnable Episodic Count for Task-Specific Intrinsic Reward
Daejin Jo, Sungwoong Kim, Daniel Wontae Nam, Taehwan Kwon, Seungeun Rho, Jongmin Kim, Donghoon Lee
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace AE-LSH hash with VQ-VAE hash in LECO (ablation) | LECO · DMLab lasertag_three_opponents_small | mean new-hash rate: 86.8% -> 71.6% | Improved |
| replace AE-LSH hash with VQ-VAE hash in LECO (ablation) | LECO · DMLab lasertag_three_opponents_large | mean new-hash rate: 68.2% -> 58.6% | Improved |
| add task-specific modulation to VQ-only episodic count (ablation) | VQ-only · MiniGrid six tasks | tasks solved | Improved |
| use VQ hash instead of heuristic DSC discretization (ablation) | episodic count · MiniGrid six tasks | tasks solved | Improved |
| use VQ hash instead of AE-LSH in LECO (ablation) | LECO · MiniGrid six tasks | task performance | Improved |
| add task-specific modulation to AE-LSH (ablation) | AE-LSH · MiniGrid | task performance | Mixed |
| remove all intrinsic reward (no-Int) (ablation) | IMPALA agent · MiniGrid six tasks | tasks solved | Hurt |
| naive combination of VQ-VAE and LIRPG (LECO-naive) instead of additive modulation (ablation) | LECO · MiniGrid and DMLab tasks | task performance | Hurt |
| lambda 0 (episodic count only) or 1.0 (modulation only) (ablation) | LECO · MiniGrid KeyCorridor and ObstructedMaze-Full | task solved: lambda=0 or lambda=1.0 vs lambda=0.5 | Hurt |
| lambda 0.25 versus 0.75 weighting (ablation) | LECO · MiniGrid | task performance: lambda=0.25 > lambda=0.75; best at lambda=0.5 | Mixed |
| VQ hash size 3x3x8 (MiniGrid) and 3x2x24 (DMLab) versus other sizes (ablation) | LECO · MiniGrid and DMLab | task performance: 3x3x8 and 3x2x24 best | Improved |
| LECO versus NovelD on Noisy-TV MultiRoom | LECO · MiniGrid MultiRoomNoisyTV-N7-S4 | frames to convergence: 3e7 vs 5e7 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- NovelD — NovelD: A Simple yet Effective Exploration Criterion (NeurIPS 2021) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Episodic Novelty Through Temporal Distance (ICLR 2025 · similarity 0.89)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Episodic Return Decomposition by Difference of Implicitly Assigned Sub-trajectory Reward (AAAI 2024 · similarity 0.86)
- Rethinking Exploration in Reinforcement Learning with Effective Metric-Based Exploration Bonus (NeurIPS 2024 · similarity 0.86)
- Abstract and Explore: A Novel Behavioral Metric with Cyclic Dynamics in Reinforcement Learning (AAAI 2024 · similarity 0.86)
- Improving Intrinsic Exploration by Creating Stationary Objectives (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
