{"id":39558,"date":"2022-12-29T18:55:12","date_gmt":"2022-12-29T10:55:12","guid":{"rendered":"https:\/\/swarma.org\/?p=39558"},"modified":"2022-12-29T18:55:12","modified_gmt":"2022-12-29T10:55:12","slug":"nat-commun-%e9%80%9f%e9%80%92%ef%bc%9a%e9%80%9a%e8%bf%87%e7%b1%bb%e7%9d%a1%e7%9c%a0%e5%9b%9e%e6%94%be%e5%85%8b%e6%9c%8d%e7%a5%9e%e7%bb%8f%e7%bd%91%e7%bb%9c%e7%9a%84%e7%81%be%e9%9a%be%e6%80%a7","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=39558","title":{"rendered":"Nat. Commun.\u901f\u9012\uff1a\u901a\u8fc7\u7c7b\u7761\u7720\u56de\u653e\u514b\u670d\u795e\u7ecf\u7f51\u7edc\u7684\u707e\u96be\u6027\u9057\u5fd8"},"content":{"rendered":"
\n
\n
\n
\n

<\/p>\n<\/section>\n

\n
\n
\n
\n
\n
\n

<\/section>\n<\/section>\n
\n
\n

\u5bfc\u8bed<\/strong><\/p>\n<\/section>\n<\/section>\n

\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n
\n

\u5f53\u795e\u7ecf\u7f51\u7edc\u5b66\u4f1a\u8fa8\u8bc6\u5c0f\u732b\uff0c\u5b83\u5c31\u5fd8\u8bb0\u4e86\u5982\u4f55\u5206\u7c7b\u5c0f\u72d7\uff0c\u8fd9\u4e00\u73b0\u8c61\u88ab\u79f0\u4e3a\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u7684\u707e\u96be\u6027\u9057\u5fd8\uff08catastrophic forgetting\uff09\u3002\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u5bb9\u6613\u906d\u53d7\u707e\u96be\u6027\u9057\u5fd8\uff1a\u5f53\u4f9d\u6b21\u5b66\u4e60\u591a\u4e2a\u4efb\u52a1\u65f6\uff0c\u5b83\u4eec\u5728\u6700\u8fd1\u7684\u4efb\u52a1\u4e2d\u8868\u73b0\u826f\u597d\uff0c\u4f46\u4ee5\u727a\u7272\u4e4b\u524d\u5b66\u4e60\u7684\u4efb\u52a1\u4e3a\u4ee3\u4ef7\u3002\u5728\u5927\u8111\u4e2d\uff0c\u7761\u7720\u901a\u8fc7\u56de\u653e\u6700\u8fd1\u548c\u8fc7\u53bb\u76f8\u4e92\u51b2\u7a81\u7684\u8bb0\u5fc6\u75d5\u8ff9\uff0c\u5728\u6301\u7eed\u5b66\u4e60\uff08incremental learning\uff09\u4e2d\u53d1\u6325\u7740\u91cd\u8981\u4f5c\u7528\u3002\u8fd1\u65e5\u53d1\u8868\u4e8e Nature Communications<\/em> \u7684\u6700\u65b0\u7814\u7a76\u6d4b\u8bd5\u4e86\u8fd9\u6837\u4e00\u4e2a\u5047\u8bbe\uff1a\u5728\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u4e2d\u5b9e\u65bd\u7c7b\u7761\u7720\u9636\u6bb5\uff0c\u53ef\u4ee5\u5728\u65b0\u7684\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u4fdd\u62a4\u65e7\u7684\u8bb0\u5fc6\uff0c\u51cf\u8f7b\u707e\u96be\u6027\u9057\u5fd8\u3002<\/span><\/strong><\/p>\n


<\/span><\/strong><\/p>\n

\u7761\u7720\u88ab\u5b9e\u65bd\u4e3a\u79bb\u7ebf\u8bad\u7ec3\uff0c\u4f7f\u7528\u5c40\u90e8\u65e0\u76d1\u7763\u7684 Hebbian \u53ef\u5851\u6027\u89c4\u5219\u548c\u566a\u58f0\u8f93\u5165\u3002\u5728\u6301\u7eed\u5b66\u4e60\u6846\u67b6\u4e0b\uff0c\u7761\u7720\u80fd\u591f\u6062\u590d\u539f\u6765\u88ab\u9057\u5fd8\u7684\u4efb\u52a1\u3002\u4ee5\u524d\u5b66\u8fc7\u7684\u8bb0\u5fc6\u5728\u7761\u7720\u671f\u95f4\u81ea\u53d1\u5730\u56de\u653e\uff0c\u5bf9\u6bcf\u4e00\u7c7b\u8f93\u5165\u5f62\u6210\u72ec\u7279\u7684\u8868\u5f81\u3002\u4e0e\u65e7\u4efb\u52a1\u76f8\u5bf9\u5e94\u7684\u8868\u5f81\u7a00\u758f\u5ea6\u548c\u795e\u7ecf\u5143\u6d3b\u52a8\u589e\u52a0\uff0c\u800c\u4e0e\u65b0\u4efb\u52a1\u76f8\u5173\u7684\u6d3b\u52a8\u51cf\u5c11\u3002\u8fd9\u9879\u7814\u7a76\u8868\u660e\uff0c\u6a21\u62df\u7761\u7720\u5f0f\u7684\u52a8\u529b\u5b66\u7684\u81ea\u53d1\u56de\u653e\u53ef\u4ee5\u51cf\u8f7b\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u4e2d\u7684\u707e\u96be\u6027\u9057\u5fd8\u3002<\/span><\/strong><\/p>\n<\/section>\n

\n


<\/p>\n<\/section>\n

\n

\u5173\u952e\u8bcd\uff1a\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\uff0c\u707e\u96be\u6027\u9057\u5fd8\uff0c\u7761\u7720\u56de\u653e<\/span><\/strong><\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\n
\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
<\/strong><\/strong><\/span><\/strong><\/section>\n<\/section>\n
\n

\u90ed\u745e\u4e1c<\/span> <\/span>| <\/strong><\/span>\u4f5c\u8005<\/strong><\/span><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/p>\n

\u9093\u4e00\u96ea<\/span> <\/span>| <\/strong><\/span>\u7f16\u8f91<\/strong><\/p>\n


<\/p>\n<\/section>\n

 <\/span><\/p>\n

\n

<\/p>\n<\/section>\n

\n
\n
\n

\u8bba\u6587\u9898\u76ee<\/span>\uff1a<\/span><\/p>\n

Sleep-like unsupervised replay reduces catastrophic forgetting in artificial neural networks<\/span><\/p>\n

\u8bba\u6587\u94fe\u63a5\uff1a<\/span><\/section>\n
https:\/\/www.nature.com\/articles\/s41467-022-34938-7<\/span><\/section>\n
<\/span><\/section>\n<\/section>\n<\/section>\n<\/blockquote>\n


<\/p>\n

<\/span><\/section>\n

<\/strong><\/span><\/section>\n

\n
\n
\n
\n
\n

<\/section>\n

<\/section>\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n
\n

1. \u5927\u8111\u4e2d\u53ef\u5851\u6027\u548c\u7a33\u5b9a\u6027\u7684\u6743\u8861<\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n
 <\/span><\/section>\n
<\/span><\/section>\n
\u4efb\u4f55\u4e00\u4e2a\u667a\u80fd\u4f53\uff0c\u90fd\u9700\u8981\u5728\u7a33\u5b9a\u6027\u548c\u53ef\u5851\u6027\u4e2d\u53d6\u5f97\u6743\u8861\u3002\u4e00\u65b9\u9762\uff0c\u795e\u7ecf\u7f51\u7edc\u5fc5\u987b\u662f\u53ef\u5851\u7684<\/strong>\uff0c\u8fd9\u6837\u7f51\u7edc\u7684\u53c2\u6570\u53ef\u4ee5\u6539\u53d8\uff0c\u4ee5\u4fbf\u51c6\u786e\u5730\u8868\u793a\u548c\u54cd\u5e94\u65b0\u7684\u4efb\u52a1\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u7f51\u7edc\u5fc5\u987b\u662f\u7a33\u5b9a\u7684<\/strong>\uff0c\u8fd9\u6837\u624d\u80fd\u4fdd\u6301\u5bf9\u65e7\u4efb\u52a1\u7684\u4e86\u89e3\u3002\u5c3d\u7ba1\u6df1\u5c42\u795e\u7ecf\u7f51\u7edc\u53ef\u4ee5\u5728\u4ece\u590d\u6742\u6e38\u620f\u5230\u56fe\u50cf\u8bc6\u522b\u7b49\u4efb\u52a1\u4e2d\u8fbe\u5230\u8d85\u4eba\u7684\u8868\u73b0\u6c34\u5e73\uff0c\u4f46\u5b83\u4eec\u5728\u7a33\u5b9a\u6027-\u53ef\u5851\u6027\u7684\u6743\u8861\u4e0a\u5904\u4e8e\u6b21\u4f18\u3002<\/span><\/section>\n
 <\/span><\/section>\n
\u5728\u52a8\u7269\u548c\u4eba\u7c7b\u4e2d\uff0c\u7761\u7720\u65f6\u4e0e\u6e05\u9192\u5b66\u4e60\u65f6\u7684\u81ea\u53d1\u795e\u7ecf\u5143\u6d3b\u52a8\u76f8\u4e92\u5173\u8054\uff0c\u8fd9\u79cd\u73b0\u8c61\u88ab\u79f0\u4e3a\u7761\u7720\u56de\u653e<\/strong>\u3002\u7761\u7720\u671f\u95f4\uff0c\u795e\u7ecf\u5143\u5728\u6ca1\u6709\u5916\u90e8\u8f93\u5165\u7684\u60c5\u51b5\u4e0b\u81ea\u53d1\u6d3b\u52a8\uff0c\u5e76\u5728\u5927\u8111\u5404\u533a\u57df\u4ea7\u751f\u590d\u6742\u7684\u540c\u6b65\u6d3b\u52a8\u6a21\u5f0f\u3002\u6700\u8fd1\u5b66\u4e60\u7684\u8bb0\u5fc6\u4e0e\u76f8\u5173\u7684\u65e7\u8bb0\u5fc6\u7684\u5728\u7761\u7720\u8fc7\u7a0b\u4e2d\u7684\u56de\u653e\uff0c\u4f7f\u7f51\u7edc\u80fd\u591f\u5f62\u6210\u6b63\u4ea4\u8bb0\u5fc6\u8868\u5f81\uff0c\u91cd\u53e0\u7684\u795e\u7ecf\u5143\u7fa4\u4f53\u5185\u7684\u7ade\u4e89\u8bb0\u5fc6\u5171\u5b58\uff0c\u4ece\u800c\u4fc3\u6210\u5c40\u90e8\u53ef\u5851\u6027\uff0c\u5141\u8bb8\u7a81\u89e6\u7684\u6539\u53d8\u53ea\u5f71\u54cd\u76f8\u5173\u7684\u8bb0\u5fc6\uff0c\u5982\u6b64\uff0c\u4eba\u8111\u5f97\u4ee5\u201c\u6e29\u6545\u800c\u77e5\u65b0\u201d\u3002<\/span><\/section>\n
 <\/span><\/section>\n
<\/span><\/section>\n

<\/strong><\/span><\/section>\n

\n
\n
\n
\n
\n

<\/section>\n

<\/section>\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n
\n

2. \u5f15\u5165\u7c7b\u7761\u7720\u56de\u653e\uff0c\u907f\u514d\u707e\u96be\u6027\u9057\u5fd8<\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n
 <\/span><\/section>\n
<\/span><\/section>\n
\u6765\u81ea\u795e\u7ecf\u79d1\u5b66\u7684\u7ed3\u679c\u8868\u660e\uff0c\u5e94\u7528\u4e8e\u795e\u7ecf\u7f51\u7edc\u7684\u7c7b\u7761\u7720\u56de\u653e\uff0c\u539f\u7406\u4e0a\u53ef\u4ee5\u51cf\u5c11\u673a\u5668\u5b66\u4e60\u6a21\u578b\u4e2d\u7684\u707e\u96be\u6027\u9057\u5fd8\u3002\u800c\u8be5\u7814\u7a76\u63d0\u51fa\u7684\u7761\u7720\u91cd\u653e\u5de9\u56fa<\/span>\uff08Sleep Replay Consolidation\uff0cSRC\uff09<\/span>\u7b97\u6cd5\u8bc1\u5b9e\u4e86\u8fd9\u4e00\u53d1\u73b0\u3002\u8be5\u7b97\u6cd5\u5148\u4f7f\u7528\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5\u8bad\u7ec3\u4e00\u4e2a\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\uff0c\u63a5\u4e0b\u6765\uff0c\u6a21\u62df\u7c7b\u4f3c\u4e8e\u7761\u7720\u7684\u81ea\u53d1\u6027\u5927\u8111\u52a8\u529b\u5b66\uff0c\u8fd0\u884c\u4e00\u6b21\u6027\u7684\u7f51\u7edc\u6a21\u62df\uff0c\u901a\u8fc7\u7f51\u7edc\u5411\u524d\u4f20\u64ad\u81ea\u53d1\u6027\u6d3b\u52a8\u3002\u63a5\u4e0b\u6765\u5bf9\u7f51\u7edc\u8fdb\u884c\u53cd\u5411\u4f20\u64ad\uff0c\u4ee5\u4fbf\u5e94\u7528\u5c40\u90e8 Hebbian \u53ef\u5851\u6027\u89c4\u5219\u6765\u4fee\u6b63\u795e\u7ecf\u5143\u6743\u91cd\u3002\u5728\u8fd0\u884c\u8fd9\u4e2a\u65e0\u76d1\u7763\u8bad\u7ec3\u9636\u6bb5\u591a\u6b21\u4e4b\u540e\uff0c\u4f7f\u7528\u5e38\u89c4\u53cd\u5411\u4f20\u64ad\u8fdb\u884c\u6d4b\u8bd5\u6216\u8fdb\u4e00\u6b65\u8bad\u7ec3\u3002<\/span><\/section>\n
 <\/span><\/section>\n
\u56fe1\u5c55\u793a\u4e86\u5728MINST\u6570\u636e\u96c6\u4e0a\uff0c\u8ba9\u795e\u7ecf\u7f51\u7edc\u5728\u6bcf\u6b21\u5b66\u5230\u4e00\u4e2a\u65b0\u7684\u5206\u7c7b\u4efb\u52a1\u540e\u7ecf\u8fc7\u4e00\u4e2a\u7c7b\u7761\u7720\u56de\u653e\uff0c\u53ef\u4ee5\u770b\u5230\u6700\u7ec8\u5404\u4e2a\u4efb\u52a1\u7684\u51c6\u786e\u7387\u90fd\u504f\u9ad8\uff0c\u56e0\u6b64\u53ef\u4ee5\u8bf4\uff0cSRC\u7b97\u6cd5\u907f\u514d\u4e86\u707e\u96be\u6027\u9057\u5fd8\u3002<\/span><\/section>\n
 <\/span><\/section>\n
<\/section>\n

\u56fe1. MINST\u6570\u636e\u96c6\u4e0a\uff0c\u5728\u8bad\u7ec3\u5b8c\u65b0\u4efb\u52a1\u540e\u52a0\u5165\u7c7b\u7761\u7720\u56de\u590d\u540e\uff0c\u5404\u4efb\u52a1\u53ca\u603b\u4efb\u52a1\u7684\u51c6\u786e\u6027<\/span><\/p>\n

 <\/span><\/section>\n
SRC \u53ef\u4ee5\u5355\u72ec\u4f7f\u7528\uff0c\u4e5f\u53ef\u4ee5\u4e0e\u5176\u5b83\u8bad\u7ec3\u65b9\u6cd5\u7ed3\u5408\u4f7f\u7528\uff0c\u4ee5\u8fdb\u4e00\u6b65\u63d0\u9ad8\u8fd9\u4e9b\u65b9\u6cd5\u7684\u57df\u6cdb\u5316\u6027\u80fd\u548c\u5bf9\u6297\u653b\u51fb\u65f6\u7684\u9c81\u68d2\u6027\uff0c\u89e3\u51b3\u8fde\u7eed\u5b66\u4e60\u95ee\u9898\u3002\u7814\u7a76\u8868\u660e\uff0c\u5b66\u4e60\u5230\u65b0\u4efb\u52a1\u540e\uff0c\u5728\u7761\u7720\u671f\u95f4\u795e\u7ecf\u5143\u7684\u81ea\u53d1\u91cd\u65b0\u6fc0\u6d3b\u53c2\u4e0e\u5c40\u90e8\u53ef\u5851\u6027\u89c4\u5219\uff0c\u53ef\u4ee5\u6062\u590d\u65e7\u4efb\u52a1\u7684\u8868\u73b0\uff0c\u4ece\u800c\u514b\u670d\u707e\u96be\u6027\u9057\u5fd8\u3002<\/span><\/section>\n
 <\/span><\/section>\n
\u56fe2\u5c55\u793a\u4e86\u5728\u5f53\u524d\u6d41\u884c\u7684\u5e94\u5bf9\u707e\u96be\u6027\u9057\u5fd8\u7684\u65b9\u6cd5\u9884\u6f14<\/span>\uff08rehearsal\uff09<\/span>\u4e4b\u4e0a\uff0c\u5f53\u65e7\u6570\u636e\u7684\u6bd4\u4f8b\u589e\u52a0\u540e\uff0c\u4f7f\u7528SRC\u7684\u6a21\u578b\u5728\u5404\u79cd\u6570\u636e\u96c6\u4e2d\u7684\u8868\u73b0\u90fd\u4f18\u4e8e\u4e0d\u4f7f\u7528\u7684\u65b9\u6cd5\uff0c\u5c24\u5176\u662f\u5927\u6570\u636e\u96c6CIFAR10\u4e0a\u7684\u5dee\u5f02\u66f4\u52a0\u660e\u663e\u3002<\/span><\/section>\n
 <\/span><\/section>\n
<\/section>\n

\u56fe2. \u5728MINST\u3001Fashion MINST\u3001Cross Model MINST\u53caCIRFAR10 \u6570\u636e\u96c6\u4e0a\uff0c\u52a0\u5165\u548c\u672a\u52a0\u5165\u7c7b\u7761\u7720\u56de\u653e\u65f6\u7684\u5206\u7c7b\u51c6\u786e\u5ea6\u5bf9\u6bd4\u3002<\/span><\/p>\n

 <\/span><\/section>\n
\u7814\u7a76\u8005\u8fd8\u8003\u5bdf\u4e86SRC\u7b97\u6cd5\u53d1\u6325\u4f5c\u7528\u7684\u673a\u5236\uff0c\u53d1\u73b0\u5728\u7c7b\u7761\u7720\u56de\u653e\u65f6\u6fc0\u6d3b\u7684\u795e\u7ecf\u5143\uff0c\u5c5e\u4e8e\u548c\u65e7\u4efb\u52a1\u76f8\u5173\u7684<\/span>\uff08\u89c1\u56fe3\uff09<\/span>\u3002\u7c7b\u7761\u7720\u56de\u653e\u4e0d\u4ec5\u901a\u8fc7\u589e\u52a0\u65e7\u4efb\u52a1\u7684\u8fde\u63a5\u6027\uff0c\u800c\u4e14\u901a\u8fc7\u51cf\u5c11\u4ee3\u8868\u6700\u65b0\u4efb\u52a1\u7684\u8fde\u63a5\u6027\u6765\u907f\u514d\u707e\u96be\u6027\u9057\u5fd8\u3002<\/strong><\/span><\/section>\n
 <\/span><\/section>\n
<\/section>\n

\u56fe3. \u6570\u5b57\u7279\u5f02\u6027\u795e\u7ecf\u5143\u5728\u7761\u7720\u671f\u95f4\u7684\u6fc0\u6d3b\u7387\u5927\u4e8e\u7b2c\u4e00(A)\u548c\u7b2c\u4e8c(B)\u9690\u5c42\u795e\u7ecf\u5143\u7684\u968f\u673a\u5b50\u96c6\u7684\u6fc0\u6d3b\u7387\u3002<\/span><\/p>\n

 <\/span><\/section>\n
\u7814\u7a76\u8fd8\u53d1\u73b0\uff0c\u5f53\u6700\u521d\u7684\u4efb\u52a1\u8868\u73b0\u4e0d\u597d\u65f6\uff0c\u901a\u8fc7\u7c7b\u7761\u7720\u56de\u653e\u8fd8\u53ef\u4ee5\u63d0\u5347\u4e4b\u524d\u4efb\u52a1\u4e0a\u7684\u6027\u80fd\uff0c\u5373\u53ef\u63d0\u9ad8\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\u3002\u5f53\u7f51\u7edc\u8bad\u7ec3\u4e0d\u8db3\uff0c\u521d\u59cb\u6027\u80fd\u4f4e\u65f6\uff0cSRC \u53ef\u4ee5\u5927\u5927\u63d0\u9ad8\u6027\u80fd\uff0c\u800c\u4e0d\u6d89\u53ca\u4efb\u4f55\u65b0\u7684\u8bad\u7ec3\u6570\u636e\u3002\u7136\u800c\uff0c\u5982\u679c\u65e7\u4efb\u52a1\u8bad\u7ec3\u826f\u597d\uff0cSRC \u4e0d\u80fd\u63d0\u9ad8\u751a\u81f3\u7565\u5fae\u964d\u4f4e\u6027\u80fd\u3002\u8fd9\u4e0e\u795e\u7ecf\u79d1\u5b66\u7684\u6570\u636e\u4e00\u81f4\uff0c\u8868\u660e\u7761\u524d\u5b66\u4e60\u8868\u73b0\u4e0e\u4eba\u7c7b\u7761\u540e\u7a0b\u5e8f\u6280\u80fd\u589e\u52a0\u4e4b\u95f4\u5448\u8d1f\u76f8\u5173\uff0c\u5373\u597d\u7684\u5b66\u4e60\u8005\u5728\u7761\u7720\u540e\u8868\u73b0\u51fa\u7684\u8868\u73b0\u589e\u52a0\u5c0f\u4e8e\u5dee\u7684\u5b66\u4e60\u8005\u3002<\/span><\/section>\n
 <\/span><\/section>\n

<\/strong><\/span><\/section>\n

\n
\n
\n
\n
\n

<\/section>\n

<\/section>\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n
\n

3. \u603b\u7ed3<\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n
 <\/span><\/section>\n
<\/span><\/section>\n
\u8be5\u7814\u7a76\u63d0\u51fa\u4e86\u4e00\u4e2a\u53d7\u5230\u5df2\u77e5\u751f\u7269\u7761\u7720\u4f5c\u7528\u542f\u53d1\u7684\u65e0\u76d1\u7763\u7761\u7720\u56de\u653e\u5de9\u56fa\u7b97\u6cd5SRC\uff0c\u53ef\u7ed3\u5408\u5176\u4ed6\u65b9\u6cd5<\/span>\uff08\u4f8b\u5982\u6b63\u5219\u5316\u6216\u9884\u6f14\uff09<\/span>\uff0c\u6062\u590d\u795e\u7ecf\u7f51\u7edc\u5728\u8bad\u7ec3\u4e60\u5f97\u65b0\u4efb\u52a1\u540e\uff0c\u65e7\u4efb\u52a1\u76f8\u5173\u7684\u8fde\u63a5\u6a21\u5f0f\u3002SRC\u7b97\u6cd5\u6a21\u62df\u751f\u7269\u7761\u7720\u56de\u653e\u7684\u57fa\u672c\u7279\u6027\uff0c\u5305\u62ec\uff1a(a) \u7761\u7720\u671f\u95f4\u795e\u7ecf\u5143\u7684\u81ea\u53d1\u6d3b\u52a8\u5bfc\u81f4\u5148\u524d\u5b66\u4e60\u7684\u6fc0\u6d3b\u6a21\u5f0f\u88ab\u56de\u653e\uff0c(b) \u65e0\u76d1\u7763\u7684 Hebbian\u53ef\u5851\u6027\u4fee\u6539\u8fde\u63a5\u77e9\u9635\uff0c\u4ee5\u589e\u52a0\u4efb\u52a1\u7279\u5f02\u6027\u8fde\u63a5\uff0c\u5e76\u4fee\u526a\u65e7\u4efb\u52a1\u7279\u5f02\u795e\u7ecf\u5143\u4e4b\u95f4\u7684\u8fc7\u5ea6\u8fde\u63a5\uff0c\u4ece\u800c\u589e\u52a0\u8868\u5f81\u7684\u7a00\u758f\u6027\uff0c\u5e76\u51cf\u5c11\u4e0d\u540c\u8f93\u5165\u7c7b\u4e4b\u95f4\u7684\u8868\u5f81\u91cd\u53e0\u3002<\/span><\/section>\n
 <\/span><\/section>\n
\u672a\u6765\u7684\u5de5\u4f5c\u53ef\u80fd\u9700\u8981\u63a2\u7d22\u66f4\u590d\u6742\u7684\u7761\u7720\u6d3b\u52a8\u6a21\u5f0f<\/span>\uff08\u4f8b\u5982\u7761\u7720\u6ce2\uff09<\/span>\u548c\u4e0d\u540c\u7761\u7720\u5468\u671f\u7684\u5b66\u4e60\u89c4\u5219\uff0c\u5982\u4f55\u5e94\u7528\u4e8e\u795e\u7ecf\u7f51\u7edc\u7684\u8bad\u7ec3\u3002<\/span><\/section>\n

<\/section>\n
 <\/span><\/section>\n
\n
\n
\n

Abstract<\/span><\/strong><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

Artificial neural networks are known to suffer from catastrophic forgetting: when learning multiple tasks sequentially, they perform well on the most recent task at the expense of previously learned tasks. In the brain, sleep is known to play an important role in incremental learning by replaying recent and old conflicting memory traces. Here we tested the hypothesis that implementing a sleep-like phase in artificial neural networks can protect old memories during new training and alleviate catastrophic forgetting. Sleep was implemented as off-line training with local unsupervised Hebbian plasticity rules and noisy input. In an incremental learning framework, sleep was able to recover old tasks that were otherwise forgotten. Previously learned memories were replayed spontaneously during sleep, forming unique representations for each class of inputs. Representational sparseness and neuronal activity corresponding to the old tasks increased while new task related activity decreased. The study suggests that spontaneous replay simulating sleep-like dynamics can alleviate catastrophic forgetting in artificial neural networks. <\/span><\/section>\n

<\/section>\n
\n
\n
\n
\n

NeuroAI\u8bfb\u4e66\u4f1a\u62db\u52df\u4e2d<\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n


<\/section>\n
\u795e\u7ecf\u79d1\u5b66\u548c\u4eba\u5de5\u667a\u80fd\u9886\u57df\u7684\u591a\u4f4d\u8457\u540d\u5b66\u8005\u8fd1\u65e5\u53d1\u8868 NeuroAI \u767d\u76ae\u4e66\u8ba4\u4e3a\uff0c\u795e\u7ecf\u79d1\u5b66\u957f\u671f\u4ee5\u6765\u4e00\u76f4\u662f\u63a8\u52a8\u4eba\u5de5\u667a\u80fd\uff08AI\uff09\u53d1\u5c55\u7684\u91cd\u8981\u9a71\u52a8\u529b\uff0cNeuroAI \u9886\u57df\u7684\u57fa\u7840\u7814\u7a76\u5c06\u63a8\u52a8\u4e0b\u4e00\u4ee3\u4eba\u5de5\u667a\u80fd\u7684\u8fdb\u7a0b\u3002\u6587\u7ae0\u53d1\u8868\u540e\u5f15\u53d1\u70ed\u8bae\uff1a\u795e\u7ecf\u79d1\u5b66\u662f\u5426\u63a8\u52a8\u4e86\u4eba\u5de5\u667a\u80fd\uff1f\u672a\u6765\u7684\u4eba\u5de5\u667a\u80fd\u662f\u5426\u9700\u8981\u795e\u7ecf\u79d1\u5b66\uff1f<\/span><\/section>\n

<\/section>\n
\u672c\u7740\u4fc3\u8fdb\u795e\u7ecf\u79d1\u5b66\u3001\u8ba1\u7b97\u673a\u79d1\u5b66\u3001\u8ba4\u77e5\u79d1\u5b66\u548c\u8111\u79d1\u5b66\u7b49\u4e0d\u540c\u9886\u57df\u7684\u5b66\u672f\u5de5\u4f5c\u8005\u7684\u4ea4\u6d41\u4e0e\u5408\u4f5c\uff0c\u96c6\u667a\u4ff1\u4e50\u90e8\u8054\u5408\u5317\u4eac\u5e08\u8303\u5927\u5b66\u67f3\u6600\u54f2\u3001\u5317\u4eac\u5927\u5b66\u9c8d\u5e73\u78ca\u548c\u660c\u5e73\u5b9e\u9a8c\u5ba4\u5415\u67c4\u6c5f\u4e09\u4f4d\u7814\u7a76\u5458\u5171\u540c\u53d1\u8d77\u4e86\u300cNeuroAI\u300d\u8bfb\u4e66\u4f1a<\/a>\uff0c\u805a\u7126\u5728\u89c6\u89c9\u3001\u8bed\u8a00\u548c\u5b66\u4e60\u9886\u57df\u4e2d\u795e\u7ecf\u79d1\u5b66\u4e0e\u4eba\u5de5\u667a\u80fd\u7684\u76f8\u5173\u7814\u7a76\uff0c\u671f\u5f85\u80fd\u591f\u67b6\u8d77\u795e\u7ecf\u79d1\u5b66\u4e0e\u4eba\u5de5\u667a\u80fd\u9886\u57df\u7684\u5408\u4f5c\u6865\u6881\uff0c\u6fc0\u53d1\u8de8\u5b66\u79d1\u7684\u5b66\u672f\u706b\u82b1\u3002<\/span><\/section>\n

<\/section>\n

<\/span><\/span><\/a><\/p>\n


<\/span><\/section>\n
\n

\u8be6\u60c5\u8bf7\u89c1\uff1a<\/span><\/p>\n

NeuroAI \u8bfb\u4e66\u4f1a\u542f\u52a8\uff1a\u63a2\u7d22\u795e\u7ecf\u79d1\u5b66\u4e0e\u4eba\u5de5\u667a\u80fd\u7684\u524d\u6cbf\u4ea4\u53c9\u9886\u57df<\/a><\/span><\/strong>
<\/span><\/strong><\/p>\n


<\/p>\n<\/section>\n


<\/section>\n
\n

\u63a8\u8350\u9605\u8bfb<\/span><\/strong><\/strong><\/p>\n