{"resolved":true,"match_type":"uuid","source_type":"original_book","book_id":"7f418071-0fdc-4a5a-a4c5-82dd1979d9c2","title":"人工智能哲学","author":null,"edition":null,"publisher":null,"year":null,"isbn":"7532740544","requested_by":"printed_page","book":{"source_type":"original_book","book_id":"7f418071-0fdc-4a5a-a4c5-82dd1979d9c2","id":"7f418071-0fdc-4a5a-a4c5-82dd1979d9c2","uuid":"7f418071-0fdc-4a5a-a4c5-82dd1979d9c2","title":"人工智能哲学","normalized_title":"人工智能哲学","author":null,"edition":null,"publisher":null,"year":null,"isbn":"7532740544","isbns":["7532740544"],"parsed":true,"ai_readable":true,"document_type":"book","page_count":501,"navigation_url":"https://workshop.closeclaw.top/ai/book/7f418071-0fdc-4a5a-a4c5-82dd1979d9c2","title_url":"https://workshop.closeclaw.top/ai/title/%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD%E5%93%B2%E5%AD%A6","fixed_page_url_template":"https://workshop.closeclaw.top/ai/book/7f418071-0fdc-4a5a-a4c5-82dd1979d9c2/page/{printed_page}","path_api_url_template":"https://workshop.closeclaw.top/api/ai/book/7f418071-0fdc-4a5a-a4c5-82dd1979d9c2/page/{printed_page}","read_url":"https://workshop.closeclaw.top/api/ai/read/7f418071-0fdc-4a5a-a4c5-82dd1979d9c2","page_url":"https://workshop.closeclaw.top/api/ai/books/7f418071-0fdc-4a5a-a4c5-82dd1979d9c2/page","search_url":"https://workshop.closeclaw.top/api/ai/books/7f418071-0fdc-4a5a-a4c5-82dd1979d9c2/search"},"page":{"printed_page":282,"pdf_page":290,"physical_index":290,"physical_index_base":0,"printed_page_status":"recognized","page_sha256":"89a792471576ebad6d5f3931a30eb8bcda268e9832ac567dc6f34d2ff6708366"},"chapter":"人工智能哲学","content":"在视觉脑皮层中，响应视网膜映像的复杂特征的单元，常常有着相当大的接受场。这常被解释为走向平移不变表述方式的第一步。然而，这些大型场的功能也可能不是实现平移不变性，而是精确指出这个特征在哪里！\n\n粗糙编码的局限性。至此，仅仅谈到粗糙编码的优点，而忽略了它有问题的一些方面。当“接受场”变得太大时，就会出现许多限制，以致造成粗糙编码策略的失败。当这种场在大小上变得相当于整个空间时，会出现一个明显的限制。这一限制一般不起什么作用，因为在接受场变得这样大之前，别的更严峻的问题就出现了。\n\n只有在必须表述的特征相对稀疏的时候，粗糙编码才是有效的。如果许多特征点拥挤在一起，每个接受场将包含许多特征，同时这些粗糙编码单元中的活动模式将不能在特征点的许多备选组合之间作出区分。(如果使这些单元具备反映落入它们接受场内的特征点的个数的整体活动水平，那么只有附近少数几个点可以被接受，而不是许多点。）因而存在着分辨率/精度的权衡比较。如果特征分得很开，因而高分辨率也并非需要，那么粗糙编码就能给特征参数以高精度。作为一种粗浅的经验，接受场的直径应当与同时出现的特征点之间的间隔具有相同的数量级。[1]\n\n人工智能哲学\n\n282","previous_page":281,"next_page":283}