åãçµã¿ã¸ã®æ¯ãè¿ã
ãä»å¹´ã¯ãé±ã«1åã¯ãªã«ãããã®æ´æ°ãããã¨ããã«ã¼ã«ã§ãã£ã¦ããã1,2æéç¨åº¦ã§æ¸ããéãªãã®ã§è¯ãããã¨ã«ãã1è¨äºã¯æ¸ãã¤ããã ã£ããã8æä¸æ¬ã®å¼ã£è¶ãã®é±ã§ã¯1åã ãè½ã¨ãã¦ãã¾ã£ããæ¬å½ã«1æéåãä½è£ãç¡ãã£ãã¨ãããã¨ã¯ãªãã®ã§ãããã¯å®å ¨ã«ææãã§ãããªãã£ãã
ããã以å¤ã§ããæããã«ãããããªãè¨äºã ãªã¨èªåã§æãããã®ã¯å¤ã ããããã®ç¿æ £ãç¡ãããã¯è¯ãã£ãã¨æãããã¾ã æ¹åããä½å°ãããã ãããæ¥å¹´ã¯ã¾ãä¸æ©é²ããããã«ãããã
ãç¸å¤ããããªã«ããã£ã¦ããããã©ãã©ãã¦ãã1å¹´ã ã£ãã以ä¸ãæ¸ãã¦ãããã®ãæãã¨ã«æ¯ãè¿ããããããã®ææã§èå³ããã£ããã¨ã示ãã¦ããã¯ãã ã
1æ
- å¼·åå¦ç¿ã«ãããèªåã®èå³ç¯å²
- Transformerç³»ä¸çã¢ãã«ææ³IRISã¨TWMã®æ¯è¼
- Bigger, Better, Fasterã®ã³ã¼ããåãã
- Bigger, Better, Faster: Human-level Atari with human-level efficiencyãèªãã ã¡ã¢
- èªæ¸ã¡ã¢ 西ç°æ´å¹³ã人ééæ©æ¢°è«ã
ã1æã¯åºæ¬çã«ã¯å¼·åå¦ç¿ã®Atari-100kãã³ããã¼ã¯ã«ãããSOTAææ³ãåãããããã¦ããããã®ãã³ããã¼ã¯ã¯æ¯è¼çè¨ç®è³æºè²§è ã«åªããé¨é¡ã®ãã®ã ã¨æãã®ã§ãã§ããã°ä»å¾ã追ãããã¦ã¿ããã
2æ
- BBFã§é£ç¶çã«2ã¤ã®ã²ã¼ã ãå®è¡ãã
- STL-10ãã¼ã¿ã§FSQã試ã
- AtCoder Heuristic Contest 030
- AHC030復ç¿
ã2æã¯ãå æã«å¼ãç¶ãå¼·åå¦ç¿ã触ã£ãããçªç¶ä¸çã¢ãã«å´ãè¦ã¦FSQã¨æ¯ããããAHC030ã«ããªãæéããããããã¦ãããAHCã«å¯¾ããã¢ããã¼ã·ã§ã³ã¯å®å®ãã¦ãããããããããã¨ãã¨åå ããããªãã¨ãããããäºãã¹ã±ã¸ã¥ã¼ã«ãææ¡ãã¦ããã°ã¢ããã¼ã·ã§ã³ä¸ããããæ°ãããã®ã§ãå¤å°æ°ãã¤ãã¦ã¿ããã
3æ
- Emergent Communication through Metropolis-Hastings Naming Game with Deep Generative Modelsãèªãã ã¡ã¢
- Gaussian Splattingã試ã
- ã¬ã¦ã¹éç¨ã®å®è£ ç·´ç¿
- Levenberg-Marquardtæ³ãããã®å®è£ ç·´ç¿
ã3æã¯æ¥åã§ä½¿ããããªæè¡ãã¤ã¾ã¿é£ããã¦ãããGaussian Splattingãã¬ã¦ã¹éç¨ãLevenberg-Marquardtæ³ããä»ã©ãã ã使ã£ã¦ãããã¨ããã¨å¾®å¦ã
4æ
- 転ä¸
- MC Digital ããã°ã©ãã³ã°ã³ã³ãã¹ã2024ï¼AtCoder Heuristic Contest 031ï¼
- small_gicpãã©ã¤ã¢ã«
- Can large language models explore in-context?ãèªãã ã¡ã¢
- Mambaã®Induction Headsåç¾
- Mambaã®å¨è¾ºç¥è(2) Parallel Scanã®éä¼æã»ç¶æ 空éã¢ãã«ã®é¢æ£å
ã4æã¯æ¥ã«Mambaãæ°ã«ãªã£ã¦èªãã ããã®å¾åºãMamba2ã¨ããå«ãã¦ã追ãããã¦è¯ãã£ãã¨æãã
5æ
- Mambaæ¢è¨ª(3)
- Tree-Structured Parzen Estimatorã®æ°ã«ãªãå¼å¤å½¢æ´ç
- Stein Variational Gradient Descentã®ãã£ããç解
- AtCoder Heuristic Contest 033
ã5æãæè¡ãããããã¤ã¾ã¿é£ããããAHCãã£ãããè¬ã®åãã§ãããæé©åææ³ãè¡ããããã°ã£ããã«å¦ãã ã®ã¯ãçµå±ãã¾ã身ã«ãªã£ã¦ããªãããã«ãæããã
6æ
- street-gaussians-nsãåãã
- street-gaussians-nsãåãã(2)
- Mamba-2èªã Section 2~5
- WayveScenes101ãã¼ã¿ã»ãããstreet-gaussians-nsã§åãã
- WayveScenes101ãã¼ã¿ã»ãããstreet-gaussians-nsã§åãã(2)
ã6æã¯ãéä¸ã§Mamba2ãæ··å ¥ãã¦ãããããã以å¤ã¯ãã£ã¨street-gaussians-nsã触ã£ã¦ãããçµæ§æãå ¥ããã®ã§ãåå¹´å¾ã«æé ãåç¾ããå¿ è¦ãçã¾ããã¨ããããããã¹ã ã¼ãºã«ã§ãããçµå±ãªã«ã«ãçµã³ã¤ãã¦ãªãã¨ããã¯æ®å¿µã§ããããããã¯ä¸ã¤ã®ææã§ã¯ãããããããªãã
7æ
- WayveScenes101ãã¼ã¿ã»ãããstreet-gaussians-nsã§åãã(3)
- AWSIM西æ°å®¿ãã¼ã¿ã§street-gaussians-nsã§åãã
- ALGO ARTIS ããã°ã©ãã³ã°ã³ã³ãã¹ã2024 å¤ï¼AtCoder Heuristic Contest 035ï¼
- glimãåãã
ã7æãç¸å¤ãããstreet-gaussians-nsã¨ãããã«glimã«ã触ã£ã¦ãããLocalization/Mappingã®äººéãããåãã§ã¯ãããããããªããã¨ã¯ããã2024å¹´æ«ã®æç¹ã§Localization/Mappingã«å¯¾ããèå³ãã©ã®ç¨åº¦ãããã¨ããã¨ãæ£ç´ããªãèã¾ã£ã¦ããã¨ã¯æããã
8æ
ã8æã¯ä¸çªã²ã©ãæã ã£ããå ¨ç¶æéãããã¦ããªããããããæãããã
9æ
- RECRUIT æ¥æ¬æ©ãã¼ããã©ã½ã³ 2024å¤ï¼AtCoder Heuristic Contest 036ï¼
- æ¢ç´¢ãã§ã¼ãº
- æ¡æ£ã¢ãã«é¢é£ã®æ¼ã
- Mastering Chess with a Transformer Modelãèªãã ã¡ã¢
- MineRLå°å ¥ãã¹ã
- MineRLå¦ç¿æ¨ç§»
ã9æã¯å®å ¨ã«è¿·èµ°ãã¦ãå²å¦æ¢æ±ï¼é¬¼ç彰夫訳ï¼ãã¨ãèªã¿åºãã¦ããï¼å®éã¯8æå¾åããããªãæéãããã¦èªãã è¦ãããããï¼ãããããèããçµæãæ¡æ£ã¢ãã«ãFlow Matchingãããã§æ°ãåãç´ããæçµçã«ã¯MineRLãããã¨ããã¨ããã«è¾¿ãçãã¦ããã®ã¯è¯ãã£ããããããªãããããããã°ãã触ã£ã¦ãããã¨ã«ãªãã
10æ
- MineRLã·ãªã¼ãºãã®3
- Flow Matchingã®å®è£ ãã¹ããªã©
- ã¢ã¼ããã¯ãã£ææ³/MineRLã§ã®FlowMatching
- MineRLã§Inventoryãéãã¿ã¤ãã³ã°ãä¸è´ããããã«ä¿®æ£
ã10æã¯Flow Matchingã«ããä¸çã¢ãã«ãMineRLã§å®è£ ããã¨ããã¨ããã«åãã¦å®è£ ãé²ãã¦ãããããç¨åº¦ææã§ããã¨ã¯æããããã£ã±ãããã ãã§ã¯ãã¨æãã¨ãããããã
11æ
- MineRL ä¸çã¢ãã«å¦ç¿ ãã®5
- Streaming Deep Reinforcement Learning Finally Worksãèªãã ã¡ã¢
- MineRL ä¸çã¢ãã«å¦ç¿ ãã®6 ããããµã¤ãºå¤æ´
- é©æ ¼åº¦ãã¬ã¼ã¹
- é©æ ¼åº¦ãã¬ã¼ã¹ï¼æ¹çï¼
- MineRLå¦ç¿ ãã®7 ãªã³ã©ã¤ã³å¦ç¿
- MineRLå¦ç¿ ãã®8 ShortcutModelã®å®è£
ã11æãMineRLããã£ã¦ãããããã¦ãStreaming Deep Reinforcement Learning Finally Worksã¨ããå¼·çãªè«æãè¦ã¤ããæã ã£ããé©æ ¼åº¦ãã¬ã¼ã¹ãå¦ã³ç´ãã¦ããªã³ã©ã¤ã³å¼·åå¦ç¿ã¸ã®èå³ãåçãã¦ãããæå ã§ãåç¾ãã¦ã¿ããâ¦â¦ããªãã§2ã¶æãçµã£ã¦ã¾ã ãªã«ãã§ãã¦ããªãã®ã ããã
12æ
- DiffusionDriveãèªãã ã¡ã¢
- MineRLå¦ç¿ ãã®9 ãããæ師ããå¦ç¿ã¸æ»ã
- ä¸çã¢ãã«ã¨å çºå ±é ¬
- VLAãªã©
- Action Value Gradientãèªã¿ã試ã
ã12æããªããªãã«è©±é¡ãæ£ããã£ã¦ãã¾ã£ãããããããã®ä¸ã§ãå çºå ±é ¬ã®ã¢ã¤ãã¢ã¯èªåã§ãæªããªãã¨æãããAction Value Gradientãç解ã¨åããã¦ã¿ããã¨ã®ä¸¡æ¹ãã¡ããã¨ãããã®ã¯è¯ãã£ãã
å 容ã«ã¤ãã¦ã®ç·æ¬
ããã¯ãèªåã¯åºæ¬çã«å¼·åå¦ç¿ã«èå³ãããã®ã ã¨æããç¹ã«ãªã³ã©ã¤ã³å¼·åå¦ç¿ãããã¦è¨ç®è³æºã¯å¤ããªãã®ã§å¿ ç¶çã«ãã¡ã¤ã³ã¯çµãããããããã¯ãè¨ç®æ©ããªãã¦ãçè«é¢ã«ã¤ãã¦ã¯é²ããããã¯ããèªãã¨ãã£ã¦ãããã¨ã¯åæãã¦ããã ããã¨ãæããããç©æ¥µçã«ä¸æ¬è¯ãéãããã¨ãæããæ¥å¹´ãã¾ãåããããªæ¯ãè¿ãè¨äºãä½ã£ãã¨ãã«ããã¼ããæµ®ãã³ä¸ãã£ã¦ãããããªè¨äºã®ãªã¹ãã«ãªã£ã¦ããã¨è¯ãã