遊戲中應用強化學習技術，目的就是要打敗人類玩家？

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"2016 年，DeepMind 公司開發的 AlphaGo 4:1 大勝韓國著名棋手李世石，成爲第一個戰勝圍棋世界冠軍的人工智能機器人，一時風頭無兩。AlphaGo 的巨大成功開啓了“人工智能元年”，也讓強化學習漸爲大衆熟悉。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"5 年後，強化學習技術發展如何？最大的瓶頸是沒法用？理想的強化學習策略是什麼樣？……帶着這些疑問，InfoQ 採訪到了西山居人工智能領域專家"},{"type":"link","attrs":{"href":"https:\/\/aicon.infoq.cn\/2021\/beijing\/presentation\/3711?utm_source=infoq&utm_medium=arti&utm_campaign=8&utm_term=0831","title":"xxx","type":null},"content":[{"type":"text","text":"黃鴻波"}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"，和他聊了聊強化學習技術在西山居的應用與實踐經驗。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"“依靠經驗可以解決的問題，理論上強化學習都能解決”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"強化學習（Reinforcement learning，RL）是人工智能算法的一個特殊分支，由環境、代理、獎勵、動作、狀態五大關鍵要素組成。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"維基百科對強化學習的定義爲：強化學習強調如何基於環境而行動，以取得最大化的預期利益。與機器學習下的另外兩種訓練方法監督學習和無監督學習不同，強化學習不需要大量的“數據餵養”，而是通過不斷嘗試使自己獲得最大獎勵。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"DeepMind 研究人員在一篇名爲"},{"type":"link","attrs":{"href":"https:\/\/www.infoq.cn\/article\/Zcq3H6vxqQwWMofZ4Qlc","title":"xxx","type":null},"content":[{"type":"text","text":"《獎勵就夠了》"}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"的論文中提到，獎勵最大化和試錯經驗足以培養出可表現與智力相關能力的行爲。由此他們得出結論，強化學習這一基於獎勵最大化理念的人工智能分支，可以引領通用人工智能的發展。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"而在此之前，“強化學習教父”Richard Sutton 更是直言：“我相信，從某種意義上講，強化學習是人工智能的未來。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"強化學習到底能解決什麼問題？它是否真的無所不能？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"在黃鴻波看來，"},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}},{"type":"strong"}],"text":"依靠人類經驗可以解決的問題，理論上，強化學習都能解決"},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"。“強化學習利用的是經驗，而不僅僅是數據。在遊戲領域，如果我可以通過經驗告訴你怎麼在遊戲中走位會更好，那麼強化學習也能夠做到。在交通領域，依靠人類經驗可以知道哪個時間段哪裏車多，哪裏車少，通過強化學習技術也可以做到這一點。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"雖然強化學習很強大，但在當前還難以實現通用人工智能。“如果你想用一個強化學習模型去解決所有的問題，我認爲至少在現階段是不太現實的。強化學習技術非常依賴算力，它對 CPU 的核數、集羣的數量、GPU 的數量要求比較高。目前，強化學習技術所面臨的最大一個問題就是算力。也許在未來，隨着算力越來越強，算力成本越來越低，那麼強化學習能解決的問題也會越來越多。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"事實上，關於強化學習發展困境的討論一直存在。今年 7 月，知乎上一個題爲“"},{"type":"link","attrs":{"href":"https:\/\/www.zhihu.com\/question\/449478247\/answer\/2001407526","title":"xxx","type":null},"content":[{"type":"text","text":"強化學習領域目前遇到的瓶頸是什麼?"}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"”的話題討論熱度頗高，南大人工智能學院教授俞揚給出了“沒法用”的答案，並隨後作出進一步解釋說明："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"實際上強化學習這個古老的研究領域 2016 前在國內一直比較冷的根源就是沒法用。研究領域大家也都清楚強化學習算法樣本利用率低，然後做出了很多改進，但是要改進到什麼程度纔能有用呢，其實根據我們的經驗有一個標準："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"零試錯：一次試錯不能有，上線即能發揮效果，還要明顯優於基線。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"offline RL 是個正確的方向，但是目前的主流研究也有很多明顯的彎路，可能發論文與做落地本身就是不同的事，大家的關心點不可能完全一致吧。說沒法用只是吐個槽，要想發論文，就只能沿着所謂的 SOTA 來改進，即使是看起來沒有希望的方向。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另外就是我們的落地越來越多，不再想着去說服別人 RL 可以用了。"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"黃鴻波認爲，之所以在一些領域存在強化學習沒法用的情況，原因在於當前無法保證強化學習模型所產生的結論是百分百正確的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"“很多領域都要求百分百精準，比如醫療領域、精密儀器製造領域等等。如果達不到完全精準，那我們就不能相信 AI，最後可能還需要人工進行復審。這意味着，AI 只是起到了一個輔助性的作用，並不能起到決定性作用，因此確實沒法用。但在另外一些領域，比如遊戲，即便 AI 出現了一點小的失誤也沒有關係。所以說，強化學習能不能落地，怎麼落地，具體還是要看領域。目前來看，遊戲無疑是強化學習技術最成功的一個落地場景。”"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"強化學習技術的落地祕笈"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"在過去，遊戲 AI 一般傾向於採用行爲樹作爲決策結構，通過引入邏輯節點減少轉換條件，迅速地組織較複雜的行爲決策，此外它的重用性很高，可以通過重組不同的節點來實現不同的行爲樹。但與此同時，行爲樹的缺點也顯而易見，比如它會讓遊戲內置的機器人看起來非常死板，靈活性不強。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"而強化學習技術恰巧能彌補這個缺點，讓機器人更加擬人、智能，提高遊戲的可玩性，同時也能提高遊戲的製作效率。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"2020 年，西山居開始在遊戲中應用強化學習技術，並組建了強化學習團隊。目前經過一年多的積累，西山居已經建立了強化學習集羣，並搭建了強化學習開發平臺和開發體系。在算法設計思路上，西山居在成熟的算法模型基礎上，加入遊戲特定的 Trick，讓遊戲整體在效果呈現上更加智能。“接下來，我們有一款對戰類型的遊戲即將上線，遊戲中的 AI 就是利用強化學習技術來做的。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"在實踐的過程中，黃鴻波發現遊戲領域的強化學習和其他領域有本質上的區別。“目前市面上的算法、模型、框架基本都不是單獨針對遊戲領域的，而是一個通用的強化學習框架，它們的特點是運行環境要與框架進行強結合，並整體打包在一起進行模型訓練。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"但在遊戲領域卻是完全相反的，尤其現在大多數遊戲都是網絡遊戲，有戰鬥系統或房間匹配系統，可能戰鬥系統單獨跑在一個服務器上，訓練系統跑在另一個服務器上。也就是說，訓練環境和戰鬥環境實際上是一個分離的狀態。這種情況下，就需要開發一箇中轉的平臺來進行交互，需要考慮的問題包括怎麼獲取環境信息、狀態信息，這個過程中還涉及到傳輸效率的問題。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"在黃鴻波看來，"},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}},{"type":"strong"}],"text":"強化學習技術的落地祕笈並不在於要把算法改得多牛，而是要將強化學習技術與深度學習、遊戲策劃相結合，形成一套完備的解決方案，並將其實現"},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"“很多人認爲我們用強化學習技術就是要把遊戲變成非常強，其實並不是。通過強化學習技術去打敗人類玩家其實是一件非常簡單的事情，並且早已得到實現。但這是在研究階段做的事情，真正落地的時候，AI 的目的並不是要打敗玩家，而是要陪玩家玩遊戲。這也是我們遊戲製作的一個核心思想。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"至於理想的強化學習策略是什麼樣，黃鴻波認爲在遊戲領域中，要讓用戶有更加豐富的遊戲體驗，而在其他領域如工業製造等，要足夠智能、靈活。“理想的強化學習能夠讓玩家在玩遊戲的過程中更加開心。在非遊戲領域，理想的強化學習能夠做出更加精準的決策。”"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"人工智能跨過“人工智障”只是時間問題"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"談到對人工智能的理解，黃鴻波認爲人工智能首先是人工戰勝智能，讓計算機學到人類已有的知識或經驗，然後再輔助人類去完成繁雜或是需要腦力計算的工作。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"近幾年，人工智能發展迅速，並逐步從學術研究過渡到產業落地。Appen Limited 發佈的第七份《人工智能與機器學習現狀年度報告》顯示，各企業 AI 預算金額較去年大幅增長 55%；同時，企業更加關注 AI 項目的實際實施。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"但與此同時，關於人工智能的質疑聲也此起彼伏，有觀點認爲當前的人工智能遠沒達到智能，甚至有些是“人工智障”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"對於這些質疑，黃鴻波認爲背後主要有兩方面原因：一方面，要想把人工智能做得更智能，需要有一個非常龐大的數據雲來做訓練；另一方面，需要有強大的算力來做支撐。比如一些智能客服、陪聊 AI 很容易出現答非所問的情況，原因就在於訓練過程中並沒有給它們喂入足夠大的語料，歸根結底，還是模型數據和算力有限。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"“人工智能跨過人工智障，我認爲只是一個時間的問題。隨着時間的推移，模型逐漸強大，算力足夠廉價，數據足夠多。這三個問題解決之後，人工智能就會逐漸成爲人們理想中的樣子。”黃鴻波說道。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"最後，對於想在人工智能領域發展的年輕人，黃鴻波也分享了一點成長建議。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"在他看來，無論是做人工智能方向的研究還是方案策劃，一定要關注它的應用價值，關注如何才能將研究真正落地到生產中。而對於還未畢業的人工智能方向人才來說，一定要提前明確自己未來的發展方向。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"“如果未來想進入企業裏，就需要知道企業需要的是什麼樣的人才，企業需要的是工程能力強、算法能力強的人才，要能夠解決實際的問題。如果未來想進入高校或研究機構，就要提高自己的學術水平。當你明確了從事的方向目標之後，再去根據目標一步步倒推，自己應該往哪個方向積累。”"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"寫在最後"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"有句話叫做“萬物皆可 AI”，在可預見的未來，AI 的觸角將伸向各個領域。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"至於未來的 AI 什麼樣，黃鴻波暢想道：“與其說未來 AI 會應用在哪些方向，倒不如說未來我們需要解決什麼問題？在未來，哪裏有問題，哪裏有痛點，哪裏就可以用 AI 來解決。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}},{"type":"strong"}],"text":"採訪嘉賓："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"黃鴻波，珠海金山網絡遊戲科技有限公司（西山居）人工智能領域專家，高級算法工程師，谷歌機器學習方向開發者專家，擁有多年軟件開發經驗，著有《TensorFlow 進階指南：基礎、算法與應用》一書。曾在格力電器股份有限公司大數據中心擔任人工智能領域專家，且在多家公司擔任過高級工程師，技術經理，技術總監等職務。"}]},{"type":"horizontalrule"},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}},{"type":"strong"}],"text":"活動推薦："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"11 月 5-6 日，AICon 全球人工智能與機器學習技術大會將落地北京國際會議中心。其中，「大規模機器學習算法及應用」專場邀請到了黃鴻波現場分享《如何利用強化學習技術提高遊戲可玩性和真實性》。本次演講將從實際業務出發，講解西山居中游戲 AI 如何與遊戲相結合，碰撞出完美的火花，點擊"},{"type":"link","attrs":{"href":"https:\/\/aicon.infoq.cn\/2021\/beijing\/presentation\/3711?utm_source=infoq&utm_medium=arti&utm_campaign=8&utm_term=0831","title":"xxx","type":null},"content":[{"type":"text","text":"詳細瞭解"}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"除此之外，本次大會還設置了 NLP 技術與應用、人工智能前沿技術、通用機器學習技術、計算機視覺實踐、推薦廣告技術與實踐、AI 工程師團隊建設與管理、認知智能的前沿探索、AI 與產業互聯網結合、大數據計算和分析、智能語音前沿技術應用、大規模預訓練模型進展、自動駕駛技術等 14 個專題。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"目前大會門票限時 8 折特惠中，購票歡迎聯繫票務小姐姐文柳：13269078023（電話同微信）"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/4b\/03\/4b0b6dd0f0aaeda4f847d54e1975ae03.jpg","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}}]}