{"id":17657,"date":"2025-05-22T10:01:21","date_gmt":"2025-05-22T03:01:21","guid":{"rendered":"https:\/\/base.vn\/blog\/?p=17657"},"modified":"2025-09-05T11:19:30","modified_gmt":"2025-09-05T04:19:30","slug":"supervised-learning","status":"publish","type":"post","link":"https:\/\/base.vn\/blog\/supervised-learning\/","title":{"rendered":"Supervised Learning l\u00e0 g\u00ec? C\u00e1ch m\u00e1y h\u1ecdc t\u1eeb d\u1eef li\u1ec7u c\u00f3 nh\u00e3n"},"content":{"rendered":"\n

Trong k\u1ef7 nguy\u00ean d\u1eef li\u1ec7u b\u00f9ng n\u1ed5, kh\u1ea3 n\u0103ng ph\u00e2n t\u00edch v\u00e0 d\u1ef1 \u0111o\u00e1n ch\u00ednh x\u00e1c l\u00e0 ch\u00eca kh\u00f3a gi\u00fap doanh nghi\u1ec7p d\u1eabn \u0111\u1ea7u. M\u1ed9t trong nh\u1eefng c\u00f4ng ngh\u1ec7 c\u1ed1t l\u00f5i l\u00e0m n\u00ean \u0111i\u1ec1u \u0111\u00f3 ch\u00ednh l\u00e0 Supervised Learning \u2013 ph\u01b0\u01a1ng ph\u00e1p h\u1ecdc m\u00e1y t\u1eeb d\u1eef li\u1ec7u c\u00f3 nh\u00e3n, gi\u00fap m\u00e1y t\u00ednh \u201ch\u1ecdc\u201d t\u1eeb qu\u00e1 kh\u1ee9 \u0111\u1ec3 d\u1ef1 \u0111o\u00e1n t\u01b0\u01a1ng lai. V\u1eady Supervised Learning l\u00e0 g\u00ec<\/strong>, ho\u1ea1t \u0111\u1ed9ng ra sao v\u00e0 doanh nghi\u1ec7p c\u00f3 th\u1ec3 \u1ee9ng d\u1ee5ng th\u1ebf n\u00e0o \u0111\u1ec3 t\u1ed1i \u01b0u v\u1eadn h\u00e0nh, t\u0103ng tr\u01b0\u1edfng th\u00f4ng minh? B\u00e0i vi\u1ebft sau t\u1eeb Base.vn<\/a> s\u1ebd cung c\u1ea5p c\u00e1i nh\u00ecn chi ti\u1ebft, d\u1ec5 hi\u1ec3u v\u00e0 th\u1ef1c ti\u1ec5n nh\u1ea5t cho c\u00e1c nh\u00e0 qu\u1ea3n tr\u1ecb hi\u1ec7n \u0111\u1ea1i.<\/p>\n\n\n\n

1. Supervised Learning l\u00e0 g\u00ec?<\/h2>\n\n\n\n

Supervised Learning (H\u1ecdc c\u00f3 gi\u00e1m s\u00e1t)<\/strong> l\u00e0 m\u1ed9t ph\u01b0\u01a1ng ph\u00e1p trong l\u0129nh v\u1ef1c h\u1ecdc m\u00e1y (Machine Learning), n\u01a1i m\u00e0 m\u00e1y t\u00ednh \u0111\u01b0\u1ee3c “d\u1ea1y” b\u1eb1ng c\u00e1ch s\u1eed d\u1ee5ng m\u1ed9t b\u1ed9 d\u1eef li\u1ec7u \u0111\u00e3 c\u00f3 s\u1eb5n c\u00e2u tr\u1ea3 l\u1eddi \u0111\u00fang. N\u00f3i c\u00e1ch kh\u00e1c, m\u1ed7i m\u1ea9u d\u1eef li\u1ec7u \u0111\u1ea7u v\u00e0o \u0111\u1ec1u \u0111\u01b0\u1ee3c g\u1eafn k\u00e8m v\u1edbi k\u1ebft qu\u1ea3 t\u01b0\u01a1ng \u1ee9ng, gi\u00fap cho m\u00f4 h\u00ecnh h\u1ecdc \u0111\u01b0\u1ee3c c\u00e1ch li\u00ean k\u1ebft gi\u1eefa \u0111\u1ea7u v\u00e0o v\u00e0 \u0111\u1ea7u ra.<\/p>\n\n\n\n

Khi hu\u1ea5n luy\u1ec7n, m\u00f4 h\u00ecnh s\u1ebd d\u1ea7n \u0111i\u1ec1u ch\u1ec9nh \u0111\u1ec3 hi\u1ec3u v\u00e0 r\u00fat ra \u0111\u01b0\u1ee3c quy lu\u1eadt t\u1eeb d\u1eef li\u1ec7u \u0111\u00e3 g\u00e1n nh\u00e3n. Qua \u0111\u00f3, n\u00f3 c\u00f3 th\u1ec3 d\u1ef1 \u0111o\u00e1n k\u1ebft qu\u1ea3 m\u1ed9t c\u00e1ch ch\u00ednh x\u00e1c khi g\u1eb7p d\u1eef li\u1ec7u m\u1edbi trong th\u1ef1c t\u1ebf. Vi\u1ec7c h\u1ecdc n\u00e0y gi\u1ed1ng nh\u01b0 vi\u1ec7c h\u1ecdc sinh l\u00e0m b\u00e0i t\u1eadp c\u00f3 \u0111\u00e1p \u00e1n \u0111\u1ec3 bi\u1ebft m\u00ecnh \u0111\u00fang sai \u1edf \u0111\u00e2u, t\u1eeb \u0111\u00f3 r\u00fat kinh nghi\u1ec7m v\u00e0 ti\u1ebfn b\u1ed9.<\/p>\n\n\n\n

Ph\u01b0\u01a1ng ph\u00e1p n\u00e0y r\u1ea5t h\u1eefu \u00edch trong nhi\u1ec1u \u1ee9ng d\u1ee5ng \u0111\u1eddi s\u1ed1ng nh\u01b0: ph\u00e1t hi\u1ec7n email r\u00e1c, nh\u1eadn di\u1ec7n h\u00ecnh \u1ea3nh, d\u1ef1 \u0111o\u00e1n gi\u00e1 c\u1ea3 th\u1ecb tr\u01b0\u1eddng,… H\u1ecdc c\u00f3 gi\u00e1m s\u00e1t gi\u00fap t\u1ea1o ra nh\u1eefng m\u00f4 h\u00ecnh th\u00f4ng minh, c\u00f3 \u0111\u1ed9 ch\u00ednh x\u00e1c cao v\u00e0 \u0111\u01b0\u1ee3c s\u1eed d\u1ee5ng r\u1ed9ng r\u00e3i trong doanh nghi\u1ec7p c\u0169ng nh\u01b0 c\u00e1c l\u0129nh v\u1ef1c c\u00f4ng ngh\u1ec7.<\/p>\n\n\n\n

V\u00ed d\u1ee5: <\/p>\n\n\n\n

Gi\u1ea3 s\u1eed b\u1ea1n c\u00f3 h\u00e0ng ngh\u00ecn email, trong \u0111\u00f3 m\u1ed9t s\u1ed1 \u0111\u00e3 \u0111\u01b0\u1ee3c \u0111\u00e1nh d\u1ea5u l\u00e0 “th\u01b0 r\u00e1c” (spam) v\u00e0 s\u1ed1 c\u00f2n l\u1ea1i l\u00e0 “th\u01b0 th\u01b0\u1eddng”. M\u00e1y t\u00ednh s\u1ebd “h\u1ecdc” t\u1eeb nh\u1eefng email \u0111\u00e3 c\u00f3 nh\u00e3n n\u00e0y \u0111\u1ec3 nh\u1eadn bi\u1ebft \u0111\u1eb7c \u0111i\u1ec3m c\u1ee7a th\u01b0 r\u00e1c (v\u00ed d\u1ee5: ch\u1ee9a t\u1eeb kh\u00f3a qu\u1ea3ng c\u00e1o, link l\u1ea1, ti\u00eau \u0111\u1ec1 gi\u1eadt g\u00e2n…). Sau khi h\u1ecdc xong, h\u1ec7 th\u1ed1ng c\u00f3 th\u1ec3 t\u1ef1 \u0111\u1ed9ng ph\u00e2n lo\u1ea1i email m\u1edbi l\u00e0 th\u01b0 r\u00e1c hay kh\u00f4ng, ngay c\u1ea3 khi ch\u01b0a t\u1eebng th\u1ea5y email \u0111\u00f3 tr\u01b0\u1edbc \u0111\u00e2y.<\/p>\n\n\n

\n
\"Supervised<\/figure><\/div>\n\n\n

2. Quy tr\u00ecnh h\u1ecdc c\u00f3 gi\u00e1m s\u00e1t g\u1ed3m nh\u1eefng b\u01b0\u1edbc n\u00e0o?<\/h2>\n\n\n\n

\u0110\u1ec3 hi\u1ec3u r\u00f5 h\u01a1n v\u1ec1 Supervised Learning, b\u1ea1n c\u00f3 th\u1ec3 h\u00ecnh dung \u0111\u00e2y l\u00e0 m\u1ed9t qu\u00e1 tr\u00ecnh “d\u1ea1y” m\u00e1y t\u00ednh h\u1ecdc t\u1eeb d\u1eef li\u1ec7u \u0111\u00e3 bi\u1ebft tr\u01b0\u1edbc k\u1ebft qu\u1ea3. Trong m\u1ed7i v\u00ed d\u1ee5 \u0111\u01b0\u1ee3c \u0111\u01b0a v\u00e0o h\u1ec7 th\u1ed1ng, th\u00f4ng tin \u0111\u1ea7u v\u00e0o lu\u00f4n \u0111i k\u00e8m v\u1edbi c\u00e2u tr\u1ea3 l\u1eddi ch\u00ednh x\u00e1c. Qua qu\u00e1 tr\u00ecnh luy\u1ec7n t\u1eadp, m\u00f4 h\u00ecnh s\u1ebd d\u1ea7n nh\u1eadn ra quy lu\u1eadt gi\u1eefa \u0111\u1ea7u v\u00e0o v\u00e0 \u0111\u1ea7u ra. Sau khi h\u1ecdc xong, m\u00e1y t\u00ednh c\u00f3 th\u1ec3 d\u1ef1 \u0111o\u00e1n k\u1ebft qu\u1ea3 cho nh\u1eefng d\u1eef li\u1ec7u m\u1edbi ch\u01b0a t\u1eebng g\u1eb7p tr\u01b0\u1edbc \u0111\u00f3.<\/p>\n\n\n\n

2.1 Chu\u1ea9n b\u1ecb d\u1eef li\u1ec7u hu\u1ea5n luy\u1ec7n c\u00f3 g\u00e1n nh\u00e3n<\/h3>\n\n\n\n

Tr\u01b0\u1edbc khi b\u1eaft \u0111\u1ea7u “d\u1ea1y” m\u00f4 h\u00ecnh, c\u00e1c chuy\u00ean gia s\u1ebd t\u1ea1o ra m\u1ed9t t\u1eadp d\u1eef li\u1ec7u m\u1eabu, trong \u0111\u00f3 m\u1ed7i m\u1ee5c \u0111\u1ec1u \u0111\u01b0\u1ee3c d\u00e1n nh\u00e3n r\u00f5 r\u00e0ng. V\u00ed d\u1ee5: n\u1ebfu mu\u1ed1n hu\u1ea5n luy\u1ec7n m\u00e1y t\u00ednh nh\u1eadn di\u1ec7n h\u00ecnh \u1ea3nh m\u00e8o v\u00e0 ch\u00f3, ta c\u1ea7n cung c\u1ea5p nhi\u1ec1u \u1ea3nh m\u00e8o, ch\u00f3 k\u00e8m theo nh\u00e3n t\u01b0\u01a1ng \u1ee9ng nh\u01b0 “\u0111\u00e2y l\u00e0 m\u00e8o”, “\u0111\u00e2y l\u00e0 ch\u00f3”. M\u00f4 h\u00ecnh s\u1ebd h\u1ecdc c\u00e1ch nh\u1eadn bi\u1ebft \u0111\u1eb7c \u0111i\u1ec3m ri\u00eang c\u1ee7a t\u1eebng lo\u00e0i.<\/p>\n\n\n\n

2.2 Hu\u1ea5n luy\u1ec7n m\u00f4 h\u00ecnh \u0111\u1ec3 t\u00ecm ra m\u1ed1i quan h\u1ec7<\/h3>\n\n\n\n

Khi b\u1eaft \u0111\u1ea7u qu\u00e1 tr\u00ecnh h\u1ecdc, m\u00f4 h\u00ecnh s\u1ebd x\u1eed l\u00fd kh\u1ed1i l\u01b0\u1ee3ng l\u1edbn d\u1eef li\u1ec7u v\u00e0 c\u1ed1 g\u1eafng r\u00fat ra c\u00e1c quy t\u1eafc chung \u0111\u1ec3 ph\u00e2n bi\u1ec7t c\u00e1c lo\u1ea1i d\u1eef li\u1ec7u kh\u00e1c nhau. M\u1ee5c ti\u00eau l\u00e0 gi\u00fap m\u00e1y t\u00ednh hi\u1ec3u \u0111\u01b0\u1ee3c m\u1ed1i li\u00ean h\u1ec7 gi\u1eefa d\u1eef li\u1ec7u \u0111\u1ea7u v\u00e0o (h\u00ecnh \u1ea3nh) v\u00e0 \u0111\u1ea7u ra (nh\u00e3n l\u00e0 m\u00e8o hay ch\u00f3).<\/p>\n\n\n\n

2.3 \u0110\u00e1nh gi\u00e1 m\u00f4 h\u00ecnh v\u1edbi d\u1eef li\u1ec7u ki\u1ec3m tra<\/h3>\n\n\n\n

Sau khi hu\u1ea5n luy\u1ec7n xong, m\u00f4 h\u00ecnh s\u1ebd \u0111\u01b0\u1ee3c ki\u1ec3m tra b\u1eb1ng m\u1ed9t t\u1eadp d\u1eef li\u1ec7u kh\u00e1c ch\u01b0a t\u1eebng th\u1ea5y tr\u01b0\u1edbc \u0111\u00f3. Vi\u1ec7c n\u00e0y gi\u00fap x\u00e1c \u0111\u1ecbnh xem m\u00f4 h\u00ecnh \u0111\u00e3 h\u1ecdc t\u1ed1t ch\u01b0a, c\u00f3 d\u1ef1 \u0111o\u00e1n ch\u00ednh x\u00e1c kh\u00f4ng. M\u1ed9t k\u1ef9 thu\u1eadt ph\u1ed5 bi\u1ebfn \u0111\u1ec3 ki\u1ec3m tra l\u00e0 x\u00e1c th\u1ef1c ch\u00e9o (cross-validation), t\u1ee9c l\u00e0 chia d\u1eef li\u1ec7u th\u00e0nh nhi\u1ec1u ph\u1ea7n \u0111\u1ec3 ki\u1ec3m tra v\u00e0 \u0111\u1ea3m b\u1ea3o m\u00f4 h\u00ecnh kh\u00f4ng ch\u1ec9 gi\u1ecfi tr\u00ean d\u1eef li\u1ec7u c\u0169 m\u00e0 c\u00f2n c\u00f3 kh\u1ea3 n\u0103ng t\u1ed5ng qu\u00e1t t\u1ed1t tr\u00ean d\u1eef li\u1ec7u m\u1edbi.<\/p>\n\n\n\n

2.4 T\u1ed1i \u01b0u h\u00f3a m\u00f4 h\u00ecnh \u0111\u1ec3 gi\u1ea3m sai s\u1ed1<\/h3>\n\n\n\n

Trong su\u1ed1t qu\u00e1 tr\u00ecnh hu\u1ea5n luy\u1ec7n, m\u00f4 h\u00ecnh s\u1ebd \u0111\u01b0\u1ee3c t\u1ed1i \u01b0u h\u00f3a li\u00ean t\u1ee5c \u0111\u1ec3 d\u1ef1 \u0111o\u00e1n ng\u00e0y c\u00e0ng ch\u00ednh x\u00e1c h\u01a1n. M\u1ed9t trong nh\u1eefng ph\u01b0\u01a1ng ph\u00e1p ph\u1ed5 bi\u1ebfn l\u00e0 thu\u1eadt to\u00e1n gi\u1ea3m \u0111\u1ed9 d\u1ed1c (Gradient Descent), m\u00f4 h\u00ecnh s\u1ebd li\u00ean t\u1ee5c \u0111i\u1ec1u ch\u1ec9nh c\u00e1c tham s\u1ed1 d\u1ef1a tr\u00ean sai s\u1ed1 gi\u1eefa d\u1ef1 \u0111o\u00e1n v\u00e0 k\u1ebft qu\u1ea3 th\u1ef1c t\u1ebf. Bi\u1ebfn th\u1ec3 nh\u01b0 Stochastic Gradient Descent (SGD) c\u0169ng th\u01b0\u1eddng \u0111\u01b0\u1ee3c s\u1eed d\u1ee5ng \u0111\u1ec3 t\u0103ng t\u1ed1c qu\u00e1 tr\u00ecnh h\u1ecdc trong c\u00e1c m\u00f4 h\u00ecnh ph\u1ee9c t\u1ea1p nh\u01b0 m\u1ea1ng n\u01a1-ron.<\/p>\n\n\n

\n
\"H\u1ecdc<\/figure><\/div>\n\n\n

2.5 S\u1eed d\u1ee5ng h\u00e0m m\u1ea5t m\u00e1t \u0111\u1ec3 \u0111o l\u01b0\u1eddng sai s\u1ed1<\/h3>\n\n\n\n

\u0110\u1ec3 bi\u1ebft m\u00f4 h\u00ecnh ho\u1ea1t \u0111\u1ed9ng t\u1ed1t \u0111\u1ebfn \u0111\u00e2u, ng\u01b0\u1eddi ta d\u00f9ng m\u1ed9t c\u00f4ng c\u1ee5 g\u1ecdi l\u00e0 h\u00e0m m\u1ea5t m\u00e1t (Loss Function) nh\u1eb1m \u0111o \u0111\u1ed9 ch\u00eanh l\u1ec7ch gi\u1eefa k\u1ebft qu\u1ea3 d\u1ef1 \u0111o\u00e1n v\u00e0 th\u1ef1c t\u1ebf. M\u1ee5c ti\u00eau l\u00e0 l\u00e0m sao \u0111\u1ec3 sai s\u1ed1 n\u00e0y c\u00e0ng nh\u1ecf c\u00e0ng t\u1ed1t. Nh\u1edd \u0111\u00f3, m\u00f4 h\u00ecnh \u0111\u01b0\u1ee3c c\u1ea3i thi\u1ec7n li\u00ean t\u1ee5c sau m\u1ed7i v\u00f2ng h\u1ecdc.<\/p>\n\n\n\n

Nh\u1edd qu\u00e1 tr\u00ecnh hu\u1ea5n luy\u1ec7n, ki\u1ec3m tra v\u00e0 t\u1ed1i \u01b0u h\u00f3a k\u1ef9 l\u01b0\u1ee1ng, c\u00e1c m\u00f4 h\u00ecnh h\u1ecdc c\u00f3 gi\u00e1m s\u00e1t ng\u00e0y c\u00e0ng th\u00f4ng minh, ch\u00ednh x\u00e1c v\u00e0 c\u00f3 th\u1ec3 \u00e1p d\u1ee5ng hi\u1ec7u qu\u1ea3 v\u00e0o nhi\u1ec1u l\u0129nh v\u1ef1c nh\u01b0 ph\u00e2n lo\u1ea1i \u1ea3nh, nh\u1eadn di\u1ec7n gi\u1ecdng n\u00f3i, d\u1ef1 \u0111o\u00e1n gi\u00e1 c\u1ea3, l\u1ecdc th\u01b0 r\u00e1c,\u2026<\/p>\n\n\n\n

\u0110\u1ecdc th\u00eam:<\/strong> Transfer Learning l\u00e0 g\u00ec? C\u00e1ch m\u00e1y h\u1ecdc t\u0103ng t\u1ed1c nh\u1edd ki\u1ebfn th\u1ee9c c\u0169<\/a><\/p>\n\n\n\n

3. \u01afu, nh\u01b0\u1ee3c \u0111i\u1ec3m c\u1ee7a ph\u01b0\u01a1ng ph\u00e1p h\u1ecdc c\u00f3 gi\u00e1m s\u00e1t <\/h2>\n\n\n\n

3.1 \u01afu \u0111i\u1ec3m <\/h3>\n\n\n\n

Supervised Learning l\u00e0 m\u1ed9t trong nh\u1eefng ph\u01b0\u01a1ng ph\u00e1p ph\u1ed5 bi\u1ebfn v\u00e0 d\u1ec5 ti\u1ebfp c\u1eadn nh\u1ea5t trong l\u0129nh v\u1ef1c h\u1ecdc m\u00e1y. Nh\u1edd nguy\u00ean l\u00fd h\u1ecdc t\u1eeb d\u1eef li\u1ec7u \u0111\u00e3 g\u1eafn nh\u00e3n, ph\u01b0\u01a1ng ph\u00e1p n\u00e0y th\u01b0\u1eddng \u0111\u01b0\u1ee3c l\u1ef1a ch\u1ecdn trong nhi\u1ec1u b\u00e0i to\u00e1n th\u1ef1c t\u1ebf. D\u01b0\u1edbi \u0111\u00e2y l\u00e0 m\u1ed9t s\u1ed1 \u01b0u \u0111i\u1ec3m n\u1ed5i b\u1eadt gi\u00fap ph\u01b0\u01a1ng ph\u00e1p h\u1ecdc c\u00f3 gi\u00e1m s\u00e1t \u0111\u01b0\u1ee3c \u01b0a chu\u1ed9ng:<\/p>\n\n\n\n