Sign In

Joint Learning of Hierarchical Neural Options and Abstract World Model

μž‘μ„±μž
  • Haebom
μΉ΄ν…Œκ³ λ¦¬
Empty

μ €μž

Wasu Top Piriyakulkij, Wolfgang Lehrach, Kevin Ellis, Kevin Murphy

πŸ’‘ κ°œμš”

λ³Έ 논문은 κΈ°μ‘΄ κΈ°μˆ λ“€μ„ μ‘°ν•©ν•˜μ—¬ μƒˆλ‘œμš΄ κΈ°μˆ μ„ μˆ˜ν–‰ν•˜λŠ” AI μ—μ΄μ „νŠΈ ꡬ좕을 λͺ©ν‘œλ‘œ ν•©λ‹ˆλ‹€. 이λ₯Ό μœ„ν•΄ 데이터 효율적으둜 계측적 μ‹ κ²½ μ˜΅μ…˜μ„ ν•™μŠ΅ν•˜λŠ” AgentOWL 방법을 μ œμ•ˆν•©λ‹ˆλ‹€. AgentOWL은 μƒνƒœμ™€ μ‹œκ°„μ„ μΆ”μƒν™”ν•˜λŠ” μ›”λ“œ λͺ¨λΈκ³Ό 계측적 μ‹ κ²½ μ˜΅μ…˜μ„ λ™μ‹œμ— ν•™μŠ΅ν•˜μ—¬, κΈ°μ‘΄ 방법둠 λŒ€λΉ„ 적은 λ°μ΄ν„°λ‘œ 더 λ§Žμ€ κΈ°μˆ μ„ ν•™μŠ΅ν•˜κ³  λ›°μ–΄λ‚œ ν•™μŠ΅ 및 μΌλ°˜ν™” λŠ₯λ ₯을 λ³΄μ—¬μ€λ‹ˆλ‹€.

πŸ”‘ μ‹œμ‚¬μ  및 ν•œκ³„

β€’
κΈ°μ‘΄ λͺ¨λΈ-프리 계측 κ°•ν™” ν•™μŠ΅ λ°©λ²•λ‘ μ˜ 높은 데이터 μš”κ΅¬λŸ‰ 문제λ₯Ό ν•΄κ²°ν•©λ‹ˆλ‹€.
β€’
μƒνƒœμ™€ μ‹œκ°„μ„ μΆ”μƒν™”ν•˜λŠ” μ›”λ“œ λͺ¨λΈκ³Ό 계측적 μ‹ κ²½ μ˜΅μ…˜μ„ 효과적으둜 ν†΅ν•©ν•˜μ—¬ μƒ˜ν”Œ νš¨μœ¨μ„±μ„ λ†’μž…λ‹ˆλ‹€.
β€’
Object-Centric Atari κ²Œμž„μ—μ„œμ˜ μ‹€ν—˜μ„ 톡해 μ œμ•ˆ λ°©λ²•λ‘ μ˜ μš°μˆ˜μ„±μ„ μž…μ¦ν–ˆμŠ΅λ‹ˆλ‹€.
β€’
ν•™μŠ΅λœ μ˜΅μ…˜μ˜ 해석 κ°€λŠ₯μ„±μ΄λ‚˜ λ³΅μž‘ν•œ ν™˜κ²½μ—μ„œμ˜ ν™•μž₯성에 λŒ€ν•œ 좔가적인 연ꡬ가 ν•„μš”ν•©λ‹ˆλ‹€.
πŸ‘
Made with Slashpage