佇 (Tī)人工 (jîn-kang)神經 (sîn-keng)網路 (bāng-lō͘)中 (tiong),每 (múi)一 (chi̍t)个 (ê)神經原 (sîn-keng-goân)共 (kā)伊的 (i-ê)輸入 (su-ji̍p)詔 (chiàu)權動 (khoân-tāng)家起來 (ka--khí-lâi)了後 (liáu-āu)傳送 (thoân-sàng)予 (hō͘)激活 (kek-oa̍h)函數 (hâm-sò͘) (activation function)。若是 (Nā-sī)神經原有 (ū)n的 (ê)輸入 x 1 , x 2 , . . . , x n {\displaystyle x_{1},x_{2},...,x_{n}} ,壓 (ah)神經元 (sîn-keng-goân)的輸出 (su-chhut)就 (chiū)是 (sī) a = g ( w 1 x 1 + w 2 x 2 + w 3 x 3 + . . . w n x n + b ) {\displaystyle a=g(w_{1}x_{1}+w_{2}x_{2}+w_{3}x_{3}+...w_{n}x_{n}+b)} ,其中 (kî-tiong) g {\displaystyle g} 就是激活函數。假使 (Ká-sú)函數g是線性 (sòaⁿ-sèng)函數 g ( z ) = z {\displaystyle g(z)=z} ,神經原就 (tiō)是做 (chò)線性回歸 (hôe-kui)抑 (ia̍h)線性分類 (hun-lūi)。通常 (Thong-siông)g是一个非線性 (hui-sòaⁿ-sèng)函數,做非線性回歸抑處理 (chhú-lí)毋視 (m̄-sī)線性通分離 (thang-hun-lī)的分類問題 (būn-tê)。若 (Nā)g是素直 (sò͘-ti̍t)佇0到 (kàu)1頁 (ia̍h) -1到1个 "s" 行 (hêng)函數,神經元的輸出數值 (sò͘-ti̍t)通 (thang)被 (pī)解說 (kái-soeh)做一个字原 (jī-goân)決定 (koat-tēng)問題。用 (Ēng)有較 (khah)大 (tōa)微分 (bî-hun)數值的激活函數親像 (chhin-chhiūⁿ)ReLU來 (lâi)代退 (tāi-thè)鮑河 (páu-hô) "s" 行激活函數會棟 (ē-tàng)訓練 (hùn-liān)閣較 (koh-khah)深 (chhim)的網路。