From aaa1a0e2ca387622eacea1843dbcef7103dacd5e Mon Sep 17 00:00:00 2001
From: NotXia <35894453+NotXia@users.noreply.github.com>
Date: Mon, 16 Oct 2023 21:12:13 +0200
Subject: [PATCH] Add SMM MLE

---
 src/ainotes.cls                               |   2 +-
 .../img/gaussian_mle_bad.png                  | Bin 0 -> 9844 bytes
 .../img/gaussian_mle_good.png                 | Bin 0 -> 9968 bytes
 .../sections/_machine_learning.tex            | 232 ++++++++++++++++--
 4 files changed, 209 insertions(+), 25 deletions(-)
 create mode 100644 src/statistical-and-mathematical-methods-for-ai/img/gaussian_mle_bad.png
 create mode 100644 src/statistical-and-mathematical-methods-for-ai/img/gaussian_mle_good.png

diff --git a/src/ainotes.cls b/src/ainotes.cls
index 7dbf32f..61a807c 100644
--- a/src/ainotes.cls
+++ b/src/ainotes.cls
@@ -63,7 +63,7 @@
 \newtheorem*{definition}{Def}
 
 \newcommand{\ubar}[1]{\text{\b{$#1$}}}
-\renewcommand{\vec}[1]{{\mathbf{#1}}}
+\renewcommand{\vec}[1]{{\bm{\mathbf{#1}}}}
 \newcommand{\nullvec}[0]{\bar{\vec{0}}}
 \newcommand{\matr}[1]{{\bm{#1}}}
 \newcommand{\prob}[1]{{\mathcal{P}({#1})}}
diff --git a/src/statistical-and-mathematical-methods-for-ai/img/gaussian_mle_bad.png b/src/statistical-and-mathematical-methods-for-ai/img/gaussian_mle_bad.png
new file mode 100644
index 0000000000000000000000000000000000000000..a01a987b5e0a198337d2552e68d37844c62dfbb0
GIT binary patch
literal 9844
zcmdUVc{tSH`}d$ivh)$7jF3WPD12s+Y)P_Y-v_0r5k?Fd%LrLR@`<u9*&9-pu`7o9
zSW;%}>tt($EMs4OXX^WXuIG>Ex}HCu>-odwI%nSJzVCBy=XLILM;RIFa2`E%6as;8
z>R!KU0)a4lLLf}}Y)8Nhul=`6;2)EhiH;_uv>WyV3=X+y7-&Es<?$T5w+@3bJLbBz
z7X-r9%J^e?A<ZQSfrzu~Uez#rWWO-#@HYK!pzJS1h=xeQ_3MUS*YponI%Ro(6SpsV
zcH4!EJvT?g!_26@=X+*O$ZTiDw8L;rRkkv}t!!(^`ZXqkd<X)2M!7*$Nlq*A(Y3?R
zC(_o=?&{WRJ^V|%d%wZYvaj3b%aFDI5`8wXZ)dCX&ES_hpUvvTw@rQ-<EuL|xVtt?
z5OoWC^T0z85iv~|L_ItV2?;*N4z_smFDe9b>1-<#<j(a-7RaY(JgkssqM}?75v~9K
zKkUh|%r|!pzA)l9n@&Nym)0%TQlq6y3o{@q;;l@B))n`RJtv<)`2`W~7vRK+5ZY+Y
zW*ef>IH-P`RuV%gu}q$CiQtCbyu5sMd9C**bbPz7I+{ChPHTDxQo~CPe)Ds<y3DHD
z9aZi&T-EDCq3&5==8w4^pBg*Z-+(R!ZPdkl30i+ose|~L^0MBUT_5(y>@C0dvoCO~
z!vSs|uuP%}1@E$Zs#{Pc>BS|b#TnbaThqx<OMAb&^h3|oM7ciwOM*IFUg}_9WyeFX
z=l*vy^|ai?Z}(DBOE}B&fUT8x{2ndgoFvta5z|0^$NDg2aG0$}L*=7cm2vAg*6;Hl
zt$o#r!5?hkWygMnBTi&)b-F~y<9EMP;+X}6MY*!Hepb1S)>aaK<flfL;rHgJyZRof
zuJve<cE&ausbl_2d{e>Yyi|zuqm|DR9ivq|dvnT_`I=Xs?d)$fIg~QBX+*NT%vMas
z?rqcWsw5hIMI!Un!oJ}5SEqG?J$a}QsV`S{9xRV81y$SOH(P|^YWY{)>>~Z%cy2K{
z!hpE{ZZvK*pv_O<HzV*?gj!PYUeeBDU6~Ej(<B~Nqo@yCGDW%Mr171G@e&K|<?q4=
z#6LgjhuvfPwzBy>-m+w0=t=-_pxjA^PQG%<AgY#_gftng_Hdvdl2g`%$$5Lmcy%Pn
z+-(`@wp8ee70eb5F;nPY*xh|lt^VE1^W>zM@YZ#@p>e<uJqIaSp@3;=hoUj1_vtFW
z79}~vUmrP2%!fL8pgBogj%5CZO9UI1W9D|ukw<Ffg_OMurB9H;!l3q#Ce1Rdj5lZT
z`?IFWC&$+E5qwRXB-wwlvd86YsRHX-!rMkbo>M;5-~_*Rl`-QH{J|cjS66<%NApmF
zF)NYhptB}2tAB-TdedAw_-i>Y>r=ZmYuCz^$lQ)XY1f4|6z)gfH0}!4jpMk^JKqsc
zotij&0mILxYm<@1ZikyT=S^ZCGlhRPgvmX;NY66&<<}^rC?IBDv@N5B#6?^0J{h$4
z8m(awP*Cw~wz_KMzvMq|tI9`m*rPb`;kOI$HoF~7yL!tR8}$V^UvKBdR=O$g#tfpV
zW^-B=Z(X|3%Jkh)f=eKV4K2Jg&|X}Sa|~n5>YERNBCK-b7P<E)xD8`JZO1fi#;geZ
z4ZqszJ|@{Y5*360ySb3vvw2{5bbJrx`T1P3*-h<o9G!-K!OsK_8SX-i+g)C6<(>8l
zt5Nk|PeWgPhn??Js^q^PWnCn*2Yb=Id$G)abiSCcd5#T`X+0eNO2wmk)Mxn3`|ID^
zb$+>`lq2u6UtFgFZ<UFcZ`$heNLun8d$iEr2~q0DRm(rQ8GMU}8eASF-t_gQ?+kn(
z7BObE__iy*a+Qzi7sw2f{gVqGc>k@7aGg%`*_Q-jeUF}t(?o}CQ$3DZKuGlWp8HY0
zC~MY@vV-lif;jWs(N7Nh(+)}V_BaExxd*2%LznyamxT{J>1HPmwlWW_3AGEOb=5X}
zq_$YW629hpKm%3*lN=6`YGZ*Lqj!U&oxUa&x~48zVXd4O3$?pYvht}rao@#d)Pbl8
zb+Z}%(My3#{?jFu{`|c#_nPUXheF#*7D5%T+%%o%%{LM=cVF)csDu!VQIK4a*2MU*
znBz&*q#vEQs$VYnb(aF6lI{l@myRBnT&kTfeb@kURfg!LD^P0Eg=Qj9<fjVSL?6M{
zGGMzmvvkZYhKH~ErtNJ7%lWn+`YmZmJO~G|>uqXc*p5m?_jy`L8GL{D%RxO3>bMq$
zbUb^enRG6A;MV-e<J2FF6lNh2xagHRUmPFPh&fLX%;RA9JAU^2OGMlar*=6XF5BBe
zDv`@I5vOs5fWl=!VeU*AY0+Z-oq7e|;Jdqe{hoJ!)@rYgy*l%(VKsf}0zRN(JyNM;
zV;qm)su|#mF{1x`|0Su;nee@T)AjbJptXe5c(zxbcTVk%2Z8L?wCmC?t$FX~nQqts
zw;bHLfVuNbh!ddbQOkz$WMawU`hD~%86N(>o^`Kp!sL$IU%^#8+W+d-TP6&Z;DSkJ
zwj7>EoR}q8o4y%ZdAxGc(S*%Yef<n^0sm{56czf#e92p7zJ4{jfBe_i+gmRRzBnKL
z^n!==j<nD4y4tS=YBRSmBd5IRe<0NYQ(p{2is&7mSl?|?B~1zqm-ud1-&q)<%2bb9
zt1?Y}CLIycD`!DuLZER!$559m=^n)r!IuZ4Sl)R}fODy-_MnmW!@Io0XIbZp%?10K
zcgU*QSMTiM_UB8~w$3X)RIdW`rQ6qko(~G@Unp?TtZF4aG4ID`S^0I%(9)|0HOFn;
z>ztG~Q5Jw*i>K7H*mqout_}-FG1->E9LxLX9Eg=T<#Z==5lc<jvsjk+-K>$Q`4b33
z|HH(jyMkj$mz8Bah*vh#Y3w4x;EN~US-JHih!G_PRbIIXOLXBk8O~5ejIn(G{ew{M
zfAIL=G@vg2eJAC6dBEucsHW5HLE6vs26y5e2o?}O;PFE$6?^4VoW01?vFXGbqzmca
zmCZ`i3ZW=GmpVKQnYB*!xnDq%hSPQgDouB>%l*DX2R$s5qgngl$3CDZ^Tp>_o0=a(
zN<z8aX5Gen*Ei+}M1_P}B$JVjCd|n7pt}8Txculf)kGp8WhmBvJ@vqh0HGN~vY@4X
zNG7?J{HG>tEZlxviHk+4)p@D|hsmnKQ9hF3Ln5uDNV2;fEtA6t0Z8e6t(L*sdT+fD
zoJDaYKOc&XNqxKB#)Oq<WvX2J=>6;l{YeZ5XDHbf0mnY&Fp^*_`Jz@!u&E@BRQP4p
zVPrb`e1*R(;+rMmSz~~W4^@EP`eC-Xpo)X{{Y_R;iU-_>$&@pkiKYppli<sMTQy~;
zRmmMIuS-4^=MlnU<N|0)m$UG52C3v}_KD=}?;AsB!8|aRk+d;Ka&6DswRasQ1(R0q
z)-OR}!Eb>239sYiBCy%nrnH$74lZ>Rz#nd{ufxb`Wb(q-Tal-|({Ee$BuM?%mSBXB
z_tHlh>$<A67s^M=hoJ}chOE4+woGjdeTh*7-HpVmwqFdz_M#6JiivU+K0)3vpohow
zoyq;@Xl$I>W97f7!8X^0^0?k3>|`5oZF}vZ+-B0jIACeUG`!*GJG7$BSNGS6ZB6Q<
zjI>UfmaNR{Pbd~)B}ehFK1HG8Ol)yNMd)3QKjU;THmM5+=cn><kjsmQjnR7O&ScwS
z^>nb@in8!~IgJC?#-u3P%H(9;4qyXWhQgAMfc%~??aw9i$KoF=0|Zz8YJ0F0P2u-4
zX%GBo=r5h(7)v`d)ykT@qvXecc^A2Kk*E5NI51U_b#W2iBiNX+DAMV(SY2I9&$p{W
zLrB7iw_7DTCY`vs!73oZI3kKrDs>@7Fo%3N?^Dw<F>}@L!Jdp|$SU8PmFN=NU{~v^
z#+?u?H1xRRKOh5MFd=B01P;=g)dRBPz<3H*$sd(kggwzKNVUS|+bae_a?gOoo{W}x
z2hzh(1^<r|OmK?4`;ADJHh|i%!6`Yp?$stL=i<%iF@L6t`!C+1t*EEJ_MuX(tV)hM
z+5y6^0izuL(o#A`#wHem6$3P1oU*Q5Be<5BH{4fb{;gm#c`=<*Oeo!(RQ~UjzB@r%
zS-T{`&{H8`cj7UP-O(FRyAvm#uW&WE#ML#y&={J0G_6SNcJM8*(F9!~)fp4+{^#2p
z29^3u0)GRgymIo-EV6K`A-7b%-Va2%Ll}7R+8J|hXnSIj@_Xf&AkU)XcFW_eayI}{
z`>V*<*4Ux%WaPtVun#XbmskYE0gWh_jvM{CvWYpA^=S&Q&eh`6@P3};p{N+a;%gYD
zTyTrg>MuDBuh!#~SzgLG^#S|frmxXq0e-Ny)~vy*l&SQbVu<U%GUkq9K>j%s`t8}f
zm-4cc&Dd0beQKc`b=(Bgg*k>7C^`!trOS#f{Ah|HdMKEj6p+cxSfcObgPi;e$;P$~
zg+R?90>Y)tC6fGA?P6+?j%kdwapK}9U)y+3T?dN{!6JnQ+GNt3F4wt0SsFFi69hx{
ztD(yG(mkz(gpv748#0OMJ40-*2Ob;7G{2HncbRXxU@e@nwg^Y$=<_JFNE9`PCT}kt
z5l{ddiMFR>JkykQO|YIbh0M7kU~vMC8=(tq%HRM(A{&`f7-1XmH5#pG&7lALlayrT
z^T49%bu+P)Vgtb80pM_=eceY`!te&3Xjp_PhI9S46w%3l{e%S96-(801@=`Fe`BES
zP#)G}Bjhh8rl-W6?9q&^Fyi2`9;GE6aSQlHQ4Gs9ppGMOb8g?oFaz}5KgBV6^v^5;
zCxFC-K;oi&V&YkIXn)D%#byx@rs{dL$(+v~VQIS6-fq4|2J{XVfkTXM@2HN13FlS4
zhI_njJPip2va`Hc<VFl4Ol`ql7$#=%0xnpJBg);^TJnz)``+#Jy)Y(5StD43A(HNs
z7rHQ<TIy5+I>(U5F&h}g@iy74AkQPc6WHzb_9T!s9QfW5XY@uln$q_Vj+kVob{Xs0
zEXR=R>Vjn=u#3swFRCjqJp>ZU;CQjJ#2;FEpN<iQ|9!!FI4~?t8GXcx4aghm6qR3P
z@PoVmJbip2TuSa9&?-;SttEq!2v^-bU*VyBt4d~pW1?JEZ|&tZP1~aldP?AMdq^mV
zXAYkSU*u6t=fXcA2&aK(9#b2LWjUf7$-;j@acXab;_=3!tKvc623skwdm%Fm5QHNM
z?VOq-Hgxv$XU=UB)1Rr+xbaTL9#Cn5#myoBU4~qh`|)5!Q1O-l0SG}M(Q6^ha^Q4+
zD^fJMXy7LE?f#FU#ZjW!o=9^5*mB5g+N6orM5?4U)J5QmoXGDFaYQoDruYriA|~zV
zv1lIFPTMjZ(aadnOX@#kFRzVCyS4e-j`inuS_sy?e0C`p8c(MmYhgVCR68SsRTv;=
zZ!}y0k$May@y_AjIi;(M+?z*Zp^%suc3u4oi%JJ7#Rd%AX$YSziDBUf`E2H#VxS1K
zU2p2HPxQ9>f2>ju)3ZZcZTC6H57Exiu@^%)tRzLbUM7>}eCe3`MP(h?3-REtv;~}B
zNkxgp+m}wxKMf~XqU`D+GkSpN1GMD^!HkY+&_>zK%I#tv+}x<(2o3_uN&}9#7Lhvo
zjy@_zG|~A6*#j}Eq>>`}uO&C4j@&On#~zwX23xQ$wqAH%>7L&$ZVoN@A^VmGSX$!i
zsoeJ>YUoCCOBdW63U$ySTnpY}5O}rys_)EdrMKG#fD7~NJ={RtMDnloRgk;&iVw#C
z^#w4v!-j`P`JIZ=iH#SU9y(Q`O`)uvfZLi3ty(!^Oox#CoxI=LFZExQ1AD2Do6v3Z
z6*^-9i7dwOIUSR-+pSBRvj$NO%-mCLrka*Zu2Di9#{pHZf-^8Ms+Q1=BR)gl*7NwI
z>KUNw?dRwlf^<~+lkHA;qdkae4R9zozSeB}R(+~cQMjlq^)qL)9Y6;T`94c;EZk_5
zGe8>*c!oaQ14;(S+`_>dl$QIlm(>2VSUzQl(!c>_>h9Uu0=Mt(Tbq-g7Dp*ZCpyEB
z6Hnt77E1fS3A^7)BH(Q+(JHqjxL&d_$mB-l_|q{QiQC&kkESNkhcV7Tw<U#?8;ZH8
zcYR&tCv!lzvcW;XS>G7KkUe}Cax(N(M?S6<$EZ{l_)8s8XLzWT(pnH5+^{$jiepkQ
zg>cLz{+YCQO?Zs-HSiTMqzE&=X3#LvT1u`Z%kgyl?Y-1`GMsx2xCk!V*eKl~C^xC2
zi~Qj_BWSKO_kdW$fgQ4ALK&8nq2xw220+pQP?MlzT9S{Fipo^f!HGY29;CfeT~3xf
zIM+*AQLgJ?%@eakn;DQOsfwm#5~*L<@oxZ+y)R#Emk%HFJ;KEg_j{dEsB`EgVDX%5
zX~3Q^l+;KmB;2}AIawUT(tQ<7|Ixa8K9^4MQbsLuLLV^JBi{iIOGUo;5Q9krIacUC
zO9gO(<Ra8)9p?DNu?FD`Gj+rN&N@Xozo$)xOJiSy<o+_2ht=B;TxCH1R~^CRZ7+dL
z=nODGfiXAYbkV*)FybIk&wX-Btd89SBjAZocv#b};fc$-f$0YB5F7}v-Z^f>m7=oM
zKLOVZT=+a~vW;;>LN#HSXdH1lx|Rsgd2z-N<H|yS3<1lZpkuyQKiC^CHczz>W7sM!
z=d}ASNau`LdyXUG0566OZqO!a8Yox~P@!Z|L<u+6Q(X=0`St12D~yjp4T+9X%!PB!
zMewi|MwFcNGJ6bsA{a+Jbt7=Ezz1vxa8p;c<IHQ^ha({J?4CUsKoXp%7;#OQLK@%%
zq&w$Y7jFgvL*~!awoM9>fhd3)e}C;doe$7ZYArHLJ2Lc$N|npacc<az@uWFY0=nKA
ztl#uw;h|=I$PemkDB;CgrGI$5Y+aS@f5?(H%X#m#DA%bA<_<54TW$xK=nMmuinMBk
z=Fv*Rh&bJaRv=;s4=X31h6t_1S$VKeGqNOcQ^~rXnnU~Xdee`03If)nrH(TnBI0an
zt9Ej9X|vq-0Dd_o#m8)-<NMS>%1#)o#dqo^kWJYSR$Yff>lv^)?wU+0rzrTIj$r9J
z)2hL8vhqc;8Ojd%J5&`C0LMFIi=ZTeuO1&V{XZiUH8me<)?gU&-b?%3D&SGuH)1to
zQi$LNXLeJ)7idGSD8T@zZ|M$q)1ufC`75&L7*Lr$`cCgVx?nLMwrclZfF-)v;P>D_
zz?p$D7hqsxTfZJO!>vP)L}_kz<B+*kE7=q_IS`DT4<C5=CgYjlvH*NCwEt-}-o3=a
zt)r~I*d~Dmq)x_0HC+=vXHnRCz&^C@$y_>Faf-<X2+2m!7S}1JM2eaJ*0D(3unB3Y
z=!_ptmV%xFNya8ty#>e&F|G6Q2@H+T1<Bs;8~P9%VAH%BzfZKOlnHe8Me%I`>45o|
z0F+sa=m7p0%)`nH_i+JZ6N1zyoiYnCpYK^pVaS&l(<l#Gc?_>cTmb2dL$R?JZ4UNQ
zUlXQ}1SlvO$N^xj_{;O0i82&)>)MffT>l-$naN&Pco6g<K}rD_oA&~?!udwa9o>bt
zKiN(H68O*zmUj($U<fUaNR^B;%dP?ZjN(jEMdQp&72}RjTc&uaj+#n0+~Po?pwz-`
zM|jh)l35%S5_l!!zQOYTjsW}tBo;TKNp>02rlQ3tWc2`@VmV;Xm+F9!3%Ha5FK%o4
zJtGHZ?*H~{+g)YM>!XQ!k)iD0B`aUSiZoZ*Jmqcyiv8dOgu+Hlo_s#n`r*(I>^~{n
zGu>0x*gaN2$F^9GA)-h_oQY|d<EzUJac117S`PC9GqY1AUzpq0`hL$q@Cb`P2NHgK
zW2RZw*0em&^6~;uSEQCotuFw<sCf%?OKSpc<tj$)mllxqKA;FG?_Jr=9R@dEa_O-|
zkIzFLz=wd;evIFME&2ZdQSXO3`E@aBiDlqkIE4I7Hkgc<9f*w*(OF$nHmMY3O$6cE
zN#g2A7_DPaLwKfq4{0S#-P6c*f$0O=T!L0&j!!v9+ux#1rpN^C$4-WP7H`#<G10k$
zo@8soe7M6)6>)07wVZsfJ*NnA<}TNiPGsoQyx%i=c9=-}$LS39fPl`wtuw8DTRi`w
zZRbDAHt{^*#ytfhQXd$Y6$L#7ZWRn=OCbI91`o3`eC7_#h?@HAUxQS7cy0^_m)xr`
z<R_z%>js<(zS)oev4_IQZ5p5e@X3gl>Y?C^&3=4>7v^}0-SgS20RXb)U-dLUFloA?
zj|r}4%-OqX>)&(y`=TBK`(aY;bU>C3Q_XRzh}cl{Tb`0C@BX&?2xfpnM8q0~qIdh}
z)vPYqRLoUCy=jU!DekZ-U^M--dIKT>hc@NVX)?6O%(?FiLSf1W7zjC&hxJt?ipa~F
zS(0H$qZ^#W=_-aU(pe8N3KXBz%RJtBUd|`5d3`MzjhtzE24KJ?9;Q++DEv*mvsY!v
ze#3rvyhTY>BHpNH7y1wcaTf{EkG!$)x<+=D;)+<b+&D0;WPS*g*rol|4-H*xdhm{;
zeRr=rX&(65DDqPb538cZ;>W$Wb+UFpKWeX9ye#vwcyj?SgncK-x&=bFjVhq1I5!4|
zh5p(WbO*R(nKfdMyLJkq1b)mn?E5(Q#Ocv(1l(K@+O(+e8NAI4rWv)jOP{*T3|$6R
z!2OlKwBNwo6vn6-6=sXo`JUtGbm~8g6{EkfwEuzAP@{4mDfufI)@k;Hu{wPJOX#@n
z;zvSlqkpi{1v7om>s@A}ze`r?LhPPK@?T9DDTrb^3@qFs!W!!H5K+3r&9*!*7o^He
zKijGi`wxOem{zJCje1V*)D!pzS-ey!*k2=kj0|p8ihm`aG?CH1o!knCwN%X{W`cnK
zQxl}{GKxmzt1w%k%}Ftescppx-O9(o8Zmu}MMs(=d>{C}IjRWAbFL|0iycMCXba>i
z9_;mgMmZ$IvVy?g9%r-{_T!tM`CpS_9--X~D#oE>)b9N$Br+<Y0`CXoj4kbZ#~%^2
zWB9QRj*Urfk8I09k%3m!RW3ya#&9~(WoFuL7H532TF+xkm_qZuIlfo@8Rf(%{|Y1`
zGon2ri`F?Azi)=N{wWxan))7KE1BDXr(KMy;w%!1t%2o!lp~d+X;m~(B;`|UnqQ}}
z3qVD=+9p6n6h0rLto@!s6k@}A{*ZCdUp(5b5V46p_B5V{Rg6EBw4wGZomM0hSC%t~
zE^c4}MHMHC!zkWiYyPAFK-SkJ8z=W@`}iAuAGB>F?o}PddUZ#?Ac4|mE>0O`*Li%3
zQD_!OxYyMq-At%=Hnqkd!t&vp=M<IBhKw*0hJ>RT_~7y8k$%@_j{=<R7+FuEuxvnP
zk#7o=PK82XPhaw|YVoUJzW%tvH<6P~oA=l(R7Sq@Tp4JpJC2DInSICdwEOpSqhs^B
z+GhF>%ME=p4`P4p;cz+&JEn-8Gssn5|Lg6u?uGeph92&%>AgoYuG`@vKBGDN1j`Q_
zA}p|{9bfim391E%R(I6yDb{e1f0$xC(Eg3<l#KC_Lr8J$-CNT|1pKI%4m^lZw$)nj
zJa8uX2{<d%se(iMtc|F}#yr}%rgu}kb0ae>aZ2Cq8m#Oi2Bdh1%TOT4*K23pvyWo*
za|UT@T0h&gdne|USdWGCuySC)otio!#{H#shd*1!J!>7^1xY;Pep@YYW8C8A{pFnr
z-zq!V&JPekihB>1ene42!9cC<o2XB|<%6LMogbLN;akAw+P;3-^hou9=S!J-(&BLO
zyqXcX*W`XTJ@rGsy+p=X;#BXApJpv1?4Fn-<++%9?M}UA3oaulHlOi{*PDAQ<WW2-
z4Y_GYH}qf>;`Xy$2DhuX8dw7VqGoBiw#)bo=Lc2Ql}+&fY}ByP_eB2s#JzpgOEl=9
zwcGAz52@2a#2_;FtIgwuxiM8z6btLWJ^aEhDpCt(5x<;ovnWN%ZItXRm1Jx~V?%R&
z;68KRwhh<bhDP(WUyYGlPfd<H(RAezX8XGyDQcE0NmNGDHo(2~v0Fad18Ia?Mi5G6
zJslw$m)41OfuD<W_=4JNqyAj|l4)P|GKi4Ig_Dl9yxC-zly#o}1a)So%j3k%wt=FJ
z_7t^Rma;K_1u82XSY$(&(6owl@Fg_;t!afk*=Mgz-`8ZZkzJmb@?o*Ce_vyT8=Q9H
z_5k=uDf;=77L}?mLz0ggsT(uav+b_B<(oNt)Bc2JXm^TzCU$IVTDc-fwmts-{lI<m
zUVLWK`+|cY4)f##k7rHV@$UN`ynFa<UJ1h!?eW60{j+IahUJfzsSi2#a9cm$$K>iI
z^P6_qTh<c;v};OJZQX0~Y$_R#z&08UGW+=U=zdk-w*fMN;F%gVq0g`SkfI{4+`6}3
z-K1J6cD#<S3`Z&XSFv-Z#vxNO<0Y=>5%rTE>Age{3KqCEO_Qfr;1BlElqxi(7_Dtu
zzxGX!lpv<_Fa7yBz35e!pHkCzMxU>C^e<&>q)>A#cR~?sXIj5L+KCgE%{ZWYM7tEI
z?Jp(A@wW%<+}N@=3`2f=9Y!=fJ~#F&#VfixZ6}4wIh$sh=G$jYamgBl6<qpCm0X*)
z%tY<Z{IrC&fAKT$q$-x3kIVF5T-zyw$7M9Vw#h{8%>^hK=H0jVpCeJ<jPK?nmp++o
z)q8K(?9L&Rj##fXT_N5b6NN`mJHb<KcEA%3+4{EuPb438nV+B;hScnD-{(82GwC^B
zhN@qglBw9VtM8Y(5jQJ&E$+;e_=)3n1CRn0zrmtF`3Cw}6z#9!j%n?}byC54ywcXi
ziYo^zYvTpe(@b_31lOKOr|88Te+kiqsjo`%{IAxF_9D=f@s9WZ*o9%F<*-YIc8qzX
zm+8qiR50G^i-obZFNb(CJ@y8sJCc8QC=`ETmb*V<_#_1fUg~3F1rZr3$iLPXhAr(t
z!vV##8oyJGmmb?+uZAYuJ=9$WFXvFG9Y%wG5+WX(jw~#aE+zhHgHTOh8|rVYvdL3j
z?UbKxF9Q$a%&_J#a3Z#it!8YTuyhE-(#_JAAFVa+EK1L^c#PH8;8=ro&w;}|a7G@~
zd$pJIWP#6P{nrOwVubhRbV!Ux-UgQDvw+CVZG`)1Ep~eQM?vOdjUcoRGubBSaY3!S
z!fUqsHh8EfrH=2;tSB)%T&{+zkKpDBpzUHxD>C$00C-^pO#q(lLcV4ty}z3&K$F5o
zy3g|DugMmY#Fgc!OKev_^TM>h=Phdd{#^XO^quj8{S|5}c<Ag<R)(yIwrU)?vOCG0
zj4;$fBNitkgyAg%?N`>zY3Yg7hb{e$>y@TVGC}%pe18rYU*4Puot_8nBT{MwLhS1H
z0135r3K6@9h>tQBG+W-?1*p|HU~k?*ZD9uOYPr~xtu?I!SpgtNd~9#gHxs-vQKKZH
zm+%KW)3w;_Nzex(o9V)<QI_nmzmiNj8VqCed=@pJnrbb$u>pECrrZ0Tgadx1yKL(6
z;HQMbSl5MokWd*NA1j(#<d0o{e`X5)EY_-V?KWaDT+qERBbwT<6zJh+E$W`f+yij;
z9cSkS+e5^{MEPu{EC8oJ>ASd%WbS^>G?Rv;Tm?$(I_9}&cQ116p{UOy<mEB2LK1$l
z^>9QGT!GOqQ-JH#j*L0a`SlkfU#WH0x!XCgziuq=hpf8zpWcU8k?frP_Wo|mp~AcB
zsxWme08O(842D6!MS0-PI!y?Y(&j03M_gPXV54qa#=niJ#||(NrwxMJzjG53k^+vm
zKR(Ln5+NBNM$PR1j*Ma381cGU4qpDLGpQa|+#*#WUAf|1Xf&g{B~C~MfP3&8U%cA>
zBl;nA)4{0Fr&oiAtK4kCC=q}$(~U8IkCQ_mQ$gzmm0>d930N*O53GP0a_5ub|BK$I
c^qkEDXq0TB<s*A3&?yGd)i%6Zs_7W=KjP!9=l}o!

literal 0
HcmV?d00001

diff --git a/src/statistical-and-mathematical-methods-for-ai/img/gaussian_mle_good.png b/src/statistical-and-mathematical-methods-for-ai/img/gaussian_mle_good.png
new file mode 100644
index 0000000000000000000000000000000000000000..71f4129ea78c100f2bf71b5510cd27799f812ad5
GIT binary patch
literal 9968
zcmdsdc|6oz^#7nxvgI)&YbX^NdLk3q*Rqc_#uC{X=E0D?&Du~=L`wD<+hdny5JMu%
zV~}MmL$XWu{dcG5`MzG?e}8}c{`vipUY|4f-gEA`=e*Ck=bjiN18vsRyr&@$2<vSf
z4I~6YkAXnwa+w(ch0FGzKf!-=zDR8~NO3p+Joq@_qN=Y7fs`h&9NaqzzEAn+-1UV(
z*jj1-=w3;%@j)PBPj73eng-e~Q*55wIDNeIbKudG#EX>YT^&g&%yU;595lH~Oru=~
zo*RktZfKn(u?RF8HNO#mCAIm<_mNfXk3iwm3pjdlvqy4PoL9Vc7zxrbCgRUecu+5d
zT)^sx886MdY>o`4H%w55DGkblBa7PwBLk}==|B5UZ78qq*Y{k%aqS8NBm-${8gc@X
zq4op;i8y_h3W130L^DA0&kM6bG9LZ^Omy{hK$bG*Ke0xC%(491{mR1jjZMfJiNvK+
zc?zTQ{TUADr2G16$6$%w`}EMQVN%`Qt%lW5spG(5r(`Dyi{h>!CU(It!O)GqrI<_a
zK8NkJ2sy6?PwfBNrXv_cGl)C&yi*)bmGvl-w61#lcL8)&?vb^JW!(Agm4HTt;Qi5w
zccH%$-(MX+nSVnT@Eh~%*Dsfc;wZJ=A=7-rzCDT-^$bR#YWzl5FiAo4Imu8%O5I#$
zT$wKXV|E~YzLGFoei9fJ(4EXRd-e$O<<9>;wg@)Cpt%oAMc$zvRetmT)I+@~wf@7)
z)f2WMQ2UxEh=?bA=o;bR^|Xn%MOKyK!9Sas$|sJGwud|Ij(#n1z2DK;nqm*H*)fK+
zh_=#IL{qB1rpdVs*#>)UQC7p`8i#pHf>#^lf`{tHUr{ja0+0Bm{nYrIWD*p&EQzDg
zOd*?sHq&V4zqTm>c0@)VMqxJl)Hc*$rN_|hvYwUZ^km8rhtm3+o*5y!NY!YD%eOYk
zCc}o|d;KAYN863c$0=tkF~YN@ZQq$9?r~5dr<M1Idt%LAHymx2NSQA?G=;UK2aMk>
zgfQxY$*q$Ikz`}{77yF?REJkijk{mHW4W8`PhpyLqM_$m6K?&|DP0(Ee#}(vHCn}O
z_*Jw&DL>95T691$c<T>C$YZd>P*^oW&TAw|uFES_lYvsh$t$9x=}Q^x>G5NcQGz9N
zD*b%O{>SS5;i1N(&4YiM<p2|Edg3TCG`RFj*+t#ge-u>A3&S-e1)4H&`g-DnkWlVq
zNan*8KPEPQR~F4{!c)U8?-j@Myg0%(vV;U0yL_5Y!tR3&zyIZbiBp=7%Ci-^8f>?A
zLYfCljd=K=KW#jT;y+|V>Q9>YcNKm#p22{MZv5e?(d!X-u+|>O-8A_vhJg{jS14=s
zptaA5{7M!(-K~6l5Tck|TeTZ4I-KTBylL02+<!@Vw<Cmdn}6WOBfEooQ%C!Vure;D
zGblYpS8}>0hr7&I+=CoUy;bCGX3X7txkH`-E0CbC2K>^0aFWM&y91zxceh@Zgk0P|
z_@Vs%5d7&GUyBd;W0kFYBfSRgPPN_rV&NQqwElj*G96*hv(!qb&Z)YrF0^N~mh_zt
zzA$M3ouo2)koy`|{7V#+L>6`J_K;y`_+TOqa^4)dtF5F7KKtG$2zyuJR%MCPkMscB
zwIrH@T{U;(oD`|C38457XC)}5Dr={V0s~ruil)hVzccfA5t;I8r1z`w{D#o_Kgq-j
zOS&`h9872QQT8dsb?h`wt{V|26Bcr!u`TCu%4-hCl9ih2_yfc`lk$%0QP?7G2(8Ed
zyu2tt;qK0fIVNE?o>bRge7FaBtB-7azsl~ijq33#*l|Lnh&$8}f$^O-xmls<zgj&j
z<A-BAS<)+RFcN_~#$JkO=A=e^h|a)%e`PW2@6)(9P!caxpjG!mDtxzdw>Hs$i`y!V
zu}AA*th!;@i)8ZpaM?Rt!h9LKI&M=~=6SfcNjh!+H`xDWC1B*ULf&=DRku_~rLOAa
z%CCNfBU1eV<jmRP#QCqGQJ6&Nd1g#g6D4eadAKKRtNdW#n_`?B+mKy&D3OiAU$8Z@
zaYOj}CSf1&NDF@^(A<^}k5J_T78NUB!>o(x^b1n7-uVP0`g|%~*nnPm%<=Hoptbkc
zA?qs7c^;8zUtx3y{h2tRM!hUH^w(6`;dbStH7X^S*%-55A*|&qi+*b`)Z=yvvo8DS
z{(&O08pStV1+QBwTZ*$ga9D#kO#-)5(TIrCg_O1o)-uRZa*}PW3Ufn%rK1ZH;)p(1
zaC|xIryg)#m`;(4zvI|0jb-C=5KxF-`v?%FIuK@$c!J1S6?sBhF`q6|so);2a5EXj
zI4#v`u8+@)JZ}Kr-5nlMu@!x@09b~_(Z_GxaYg_>48PYo<EDppZeF{uy!ZL5=94bO
zpSUCMlF((3^~QjDktc0%>c+!ENQ(njyy=VxqOI6^BR38LgGocasQkvxc02^aeW8^O
zV)~qe31XzG#t%_Bc?yL3|4rO$r4yZc+7eanA~^Qv?w&r@bAotq^KApx20KeI8;;dq
z?RZb8^^}9j)b>Pm!}b9}q3U>^nvTtT`@ngttUx(MvSFrmxDaA?5o7|~bzZ@%N0_SB
zWB@L2$9{Fn&D>+lSO5xmDblpOmoQrXc(e+->gPSRFfd$zuFnC$F9zd_xDzW-^YKKD
zkL!=_rM0ie=C|lf6VnkHe{CK9I$DwnpxnKcei=L5eb~fQezG&!MH{R5W~h@S;q+Zw
zIaiA<|0*qZh}L19!s|D4r>hP5sMJrf1>=d5jUkR}2K{;;R?XzbZ=E}UUwsE5v|z-I
z&OHzq3m{UUlWKVo81-;nDOi2YPo$loVg^E`(^BvxY}Wu)P=gt{(j<G;b~|#W&n>KF
z^WA%Zt^>kgpFV!a#14*t@HMiZmdvk)g$7c8S0*Feeb#EZ;m68B3A@on5a>>Yr=PLH
z%h#9*?OGm{T64}IvBy8E<G591nK3H!0^jz0enq<RC$t%idA_akd=PdpB0dt|I>D47
z4ivUQ#kv)o?ChT3<Eb<|sNdNw+8PfID5bxwEX;PfQe7cnrtW^xYwsEd^Y;Ay@4lwW
zy-Gn6aQ$q!0VL@K2UF6>@v_gGsr%K&hGOV~T<r9_z2<Ge`Xkt&8>!gRy<#wg0o4+#
zzuaD0y67UumQN4XEi$O|DUBOs!CVum<V)V$r%%FxrH~t5x)i8QP{-cQp-(y4*lDU*
zArYf843@g0sBkVD89GH(Q<w4|!RZQ2(<jA%DgRt&M_)15@09agnunuWsFlm@<-w~r
zVp%TB0XFFfEk&9Y;oWXKY`IBg?|mh~?Fr4&DtBp$$^O(o>@oDcAG#-u-A1;o>H9#w
z=tVQ!xrlAbZ1yfN{k)Jvwo_i9q5vHv33o2KPa$EvP@6#GHL?(kh27jE2BRxL{o;7h
zt}q){*H^W>3xH$}NW9ms2+LlQ^$)HP%4D)B_{Qy*efqby+8%l>*ZHtmlytk~u&FI=
zktRw8j#>}QBklpL`A26na&aKK2%UqSuHdEn0ustQ!tqFxhN;ZT&*yX0vP8`?o`EC#
z1F4NaTw)!%o%aqC4(!0;q!AlndK#Za)`Zzw7BWhV;41#vfnDYU;9OhApL4@c9?3lu
z@h1&%B|44W)$>HnaN!UC{q6_f?yz11ej_n8erZ!Dg*#q4xZS^Wxi52p>4Y+Y-_t|Q
zdk`5H0NUn%2}~M!UYk?fv%&bA;QA#fX(VG!%E%c^mzfU#W?Y!+V6gCa0+Nra<KJB7
z8=EIqZTj#}4#51}M4uoPntUHtEG(-lyV=7k?FuISfw}s?)g_q6Dk)!`FZVlN@nv)U
zQZCr>0i*?>iuyI8W3V+x>~tsD>CwFfzifgFzqBW9nl^S-++5#@#fLR+sLIYqXIX^)
z%u4{lnuJOnf|T_+o@jQO($Aps21PK}E1x+8sq+^Y4B>Ox?Tx@rCXq#&&|+Jg)IpIV
zk*1uWHXIA4i37lJQ;wBi(z3aG!OROn*uNwdfB8r~8^OzoWCIjGi-+vLSZ@j#kw3kv
z6f9OP;LPv@Yi1xt3!n`_{-xq_MOwcV{*ZU#V}qsat4d{*#Q%`+Rvs_R``ok5_*sQ0
zjU|EfJp4e_DJ?NqSUD%Zj96uxkx4o9E(_5-853Ypd!%-P89vHG=WspODi^Ym2KLnT
zf?^0Q$R&_lWV4SR*q|Y_bdXWRj2+gVh?g(_GcF8+=R~coGD+(IRdL{rt&Qyld7}yE
zZ$6R$)Lugx)r`M#;P!%!nGOH%bKt-S)es_Uz!t}hWFN8cTo7iPt25a2Db13-abs{x
zyl%(E5H~4bR-l+qzX4+1h7G~C(7hUIoulQ0tIIik2Ead!w$v4?=QekNTWwp>C%lw?
z4j{=~K|Yu6rm^*39XTXIsyuEI31d1F3x=DKV=5Wb@~;$p$uBY>ljvH+KvMaNz*P{W
zh<aR$uulQ@Ucz=s>|C~$pXrPL2;TAoEs`jkq7U2Y`Ed-9x1$;MoI~2Ts#LhLstC61
zLm6;39x=c)O|)(_0ojAPy9|z)wzvpbJdqHk5WHOLAnWAdkhyGj5no$FZj@6g0VreV
zM~I4>CL6yb#9i0s<~aeDh%*hZF(yQL7y)@{TP96uD2_eyea&6amkGFl1br@!Tu(Nh
zXCGR4W>x|$cYzua2Ar9QD62a}U`+Q~D(^fs@yDn%0tP(-!<7=h1Do5I%U!z58bI@B
z)cAXSr^9vh-FR>h@U#sv>PIsK!OhqMoTCOs9336KO2ldEN(Jmu1i_VOoN44<aWh!*
zLLyVI1ek4-zA%N<$7tUsz>7^<93yOjtG&B3(4v_*!F)3d-M6rnk-Ct~)XPE}+Ywa5
zsSI4)W?Uy!5fTKfCP<tjgs$+;y+LCQbQv@|cNFTi>2r@9IUaCUyRE3b%@_vre*|d&
zXQP;SS|SUv5hm6lI-rq$zA1#ZtJZwHd7Q0cYF3r}?adYw7}LgMAZyo7_cDoqhj03@
z;==-IYN5#$JyP<K+@i0iSJ6PYI>0VPykRWG+^Y*j<C<0&37A?iXNOT8T0RXdYk$UP
zIY;c$CLvzBx?3k7`4l0K$I<!pzXf^0yJcG)R1-;V+<sWZ4wHx&2O}RW%81JN7g{o<
zFHor`Y1)zY22p$%YUIEZLRpyNMYjBWOhIHM1GgGSs*m?r?#4Qp83O!Bn3SK56z^51
z&A>ZX;csU8zkEBEVaNY3Y5v1DWuOIz!&NqDsS@DbtY*L0uoX$ExUq{%{kwJ$IN}0T
z-UdRK!XnI;l1%p5uC0Q)Rl54`qfvw~4tqFYX^Y(S!3+QNPb!SNenx1HlQz&)5#`Pt
zB9H?QKp311(u`)9Ei_R?5%L@3sUDmq^O^8&>5vQ<nB4vHaLT;Modxa>wM#IMoMLD=
z-%98ETAsIXv3K79ijQO1(*UzQLw;;I49qw$#2_~x0<F`m;bu&2@nBHRb9tr0a)Xwi
z10=e2kVg5wOjY8bD_+qLv(^mFs6}K50WJpd7vQp2gZu-ss&dm(Qe1;^spKINGu)m?
z`w`hPHqerjLxeL`DWM7>vs%#%&R*_}bKO+@ZAFw-UTsYvUDqoPrmm5l_$_Pv!jybm
zZO!1lovF{oTW&B2<t#PguI@m~JiA;u$zxR8Cnr9et`>+=jzt(Jl?bAp@g;@}i_E=}
zKr@x5<s(lET*9wc8X)~KsYmqdASEh4TXfBe<yY3MuHsl1Ag}q8-A|-QfZ?y!lsBp>
zKY#mXIqG35X+<G|#3G2^PaQ~jQMUe2uSm1@rF*jjkCL<nIAaNIY;`rsat5hx@!GbG
zXekTSHw&VCEp*emNPRVHu=6W<rFBT$wqMic?Y*5Q6v!Bckek-IGF|cac2r7epD#Ck
zQV+WCEX>^nZVUq!TxqKKHj6xbQ<O6ZsMr^XYZFsjL{q%GgRV<@<WxZ%gEP2!13~Wa
zZQ2w)-RrtawEgdlinG<)GA;NjWnuV|v<}uLPvBV=Z2@R0bNz<=sDF>5mOQlFG#CZ)
ztAHnn$?QyL2{ZP*hR<Wbb7`Jz;2xsA4@aFqsIUGITcBD+O+|qO4k{6m2mmQjzO1gn
zRyVw6_ip_bD~T|l0<N7<+A?gHVple!(%zGYyeAm8fhW<1pVKTdC!AaUG=R?S*8Y92
zzg6i6()KEQ&P~+9)NG!K&DaZ@Ws_?VB(M#wjt|=scE*J%38pq<HJm6Fnvc<lik^~x
z(&7gIRrAWIDbqt?6@dI*mt{A?Ug)wamIm7F8vqrnR0F8b;yF1mG9%~yaZGO+XKI7W
zi@N|RqS1k7Aw$-2=G7BO#kX^vOugW&<QD=I`R9$^?$t6>0K<w0&gGIq?ll1l{RqCZ
zfoJK+;+<A9;c(h9qCyVooIwP*t<B|o%W*E$W77WUr8XfZNF!qY3F7?xWK|J~;MP}w
z8(D}lP5)4fn7j|jZvVNHdwBNV3NR7_VDcjx`sM^dc)ZDc19Lkqe!RG$0uZkPq&Gws
z7+tZD`syVyu@PJUq|=}AqFxaoO1S!7=<_sJM1|+Xn*!Uo4pOljJ-2FdfZ;_gmY5zZ
z`eT~(z+s$Nn2Hn~2sZ>41=24>^ZCUo*Vs5#u%%(($6Ryla<}j$<sGalm(=)u=ntnN
zFKzlO{k=KmaMDKvj5kmm^8rFS1FMhc(^4Fxc|$>QCLH}c56k`68bD~I4g4u-yC3N=
zkUhZs9B8E-AH}Bx!W1}#^Lc{B{2*M?y!Gvo&vy_`=U#I#&5&DOMFEuHQ{#Vb!fqi5
zWI{7zB7$mO8pD`q#sPkRc9WnY_XMG$h&q)+JepPjtLUroX9yNQLE42U^U)NY|MA|?
zzxWSg>f5&*OtOI8TOi`p!j`BRkN{|sMyiKi{H-chHGcBjpqgJ+#Hm;yEu77jgPPG6
zzcliyFiLm8@me4zo0kTY@GuGltX{gNT(P>KAz<#|V;=}D@epc$wdc>UpdI~~S9ic7
zXW?(2cjUgRS_G&E7&+qi#D$AdNCgbo%lTGSeNB$&J;HqMh}D@Jt-rC}HP$W>T#l~1
z%WW-5JI@5>m`ulTOq#pj1PBw>cVTmpi44*L*;l9O@L39xQryMtbstVwUTRfk)fZ@A
zzWOc|$wkjYG_8pj`&!gKrP2UaP}lM@(<Q*+{NQ3%O2423B>U67AP8FGugi@9`Ks)F
zviW9qSm;u#YQrgv(e-pKFA_%l`zzk3P+5%j_Nv6Yh+0l+hBHz$4QPDAl)Z6Fbd}tP
zFk%Oq(sRGxPpw>x<#wvtc)3X;bTGs7rG(ksXFso8iZe?zwjRhBV3tk?T+TlIS+wcX
z!c!?rX>HXQjgtY{DeZV(fTXGJpzf8Gq!kWAQ~u|F?^wE0A1f@4PyR=IGANQoOL|qK
zAE)+O-IsBdrI!JF<Fbl-7H?X>sZB6a22)k|t8-!}Of-Kb*%1ale@IuT<b9Y5WM^y(
z@buni6w(F}f^fTllXe&kJ-2q!90n@p2A2KJOry+2BY=?4Q~xn~edI$+f}{)#3OH4w
zA9KR$uH&z;;?22Dwe!qHvC|n!PY^tq=@{>Y(l{>F_=Q-zM$@cWa^FqqRN4=@GSA>Z
z9B(!LXVNCTTGoVt=y_Hnq5t{h$G=N={dQ0&YL<Uem`%%a!S!oWWZ%mX+~{`GA?Y-a
z6Bw9}xPyZq6}`@n;=2-W`T;~j9^-iAqHw&aiEKP0wPlWrYOitgPEo@5Wbdoy`k*v%
zyOdrFghA2E@&6G1N|L{Jt#8W#%a2!6JcW^dgw%G6M`~N}<8v9Q3dpB@Vy(^0e<fKd
z>q_0VD@>I+`b~+(|6?rPP#)64O(I+g6GDJ-7h9X7F-ca2K!h+L!Zjd*$J)pi3mY2$
zY3?l5{)wOb9sN*WmyN<2P$?jENSg830O_9_XyftDe?NO?piS7vO+9*X0r109jTdO(
z5cko#6dx+La*ko`-UT3>NY|Z<e(nc+jmi;yB7h&_@1jr`aLoa9&Y&JY9<G04JG2qp
zX0x=wNpL7g(l9L0{m$|i={;yM@o!66w*!_e;B1`I;MRuqJnjdDSiHcSE%?zhBH-^n
zZC}BRk)}sZymgcld&IxlKd&=@=6?niW)p{FE!mdGjBm;25MxY)TrQz6YP0YvPTTtd
z<G>iRz*Nz-bKJRt#pBeW2_86)mm0q*Y9U%rlE3-CFU!1fpEgfUdEC?;fkuV_(N@)`
zIYBMw$~DzdKGC)hi^hNsftPEN;=d4E^}5OuhTUapCzBa~)L!HS_xy+jE^0>Rmg3L2
zDi+q#^)j0Bx%4PVY%`r!A_o(<|M@ljz}-k(P)lFiNhThEY^FwII{K_K0UsVF8Nm3-
zZK~i6#7GtIP3#*Rc*=;>^NHxY(5gymWkV6(bv$8oez|ve?-7G3xWRGv<BOtCv36Z>
zp3j9N4mRJ_#c><Sf)zjKS8(4$n0vaH;QG-g2)faaPoE=I{k{iC?Pc}fGONQF(Kxq~
zy*^2jOiak~ZW3y8V2uj7V;l=xQ8)tf#W4DDY_v^{<Nka2yIjA$+%xIfSTBV`Iyzce
zL_DKkZWtlhZ~kO>j*Qk=x62m<#Pjs1ML5bbY`4wfBqexwKnNJJNw`t(0gRv#{rF{^
zcQGbu<c_c1ETJDgX1B5zR!UC*-uN;ubDazy3m>+J#edvS*0<m;5CZ~As_`dT5?Qu8
zw}?KaEU>TE_lgayt8N{KlE0w&O`|xNE<)S$@v7U{skx`gHlg~Z?y<^<rH1y4OtkvY
z9d`n>8_s{oRfoZk)a-=1#4;qk2XiX7SA>25Gq|jCT#`0!a;y2kzZUQQ-;aK-KjQB{
ze6W=IYy?M+iHWI#J?09RwtP`DW1lPjd-_KB%I@vIU}%!rGDFurqC>2@d+x%@@d37!
z9@JVMzl?b;5`|p>nEl%OTME<7xk@~xW=r{e>uSr+A#)NifQv6D&eT^oK(AnLBd<$+
zu%PA>`_ZVyVrg3aB}oV6tW)_%jLq3PpN&yMLRh_}tY*aMrcgD^P~{D<;j<XeN?O6v
zP`?j(VStf7NYZL3n-?N9Ry9y6o&@gB7klS<WU&(!f82X+ez=`*DInD5uZUj&E>Fk{
zX4!5Es;`{OXZDd$-YcczuN-Bp?aYRQSO5wu_Me^QwQRz6=Msi^`S?77uc6|G8icKD
zg8bLWGnU+6G@~C2$2t@6j^ge4XGV92juRaGGcTd{w`Yp4Ne8O&pSjI%NRn$c-Dd>9
zTOX)W>wCtHZl%wyjS{U2Bkloc(aH2uch}B>x92C>@fYNx3j{}x{kGbzy~W=}+~cI$
zGt(-l^~8KL8Ef3=LnG)9yW@{D#Xfqz(G%@WKY_>FU$YEZr6^x{mrC6cPc6zFQD~C(
zmp4D`v<aAPeE)C?Y$AfnH%>FGGB)^#yhgd%<+T3Sl0CNm(_cybF;s{6^Gq$mt<4b{
z9+3ug=-uypzxG5g7wj--<#h;i&iqCWy2k*rAgyvW29H~(`KCc>*7}y}0}>jOcYYhl
z3Z1cUdV-iM|MoD>U>gsCv6|UoH<<>SjnA31A1&h|)KA<@7O?ke6BFb1$Y|dhUr+3?
z*w-U-$Q=P8BrDl{@TXXhs!hZC94?^qa|^%u4b144yUzia+B=V>WF8fnG*?5~FYmM*
z*MFGai2PO*H!-bp%#7)ET+{a&0(IcEv@}S2ezbDoE#hv0K)?)jdykM1a&_2+Ed|s_
zf-Vq0EH~`8@2*%AjwCg<_g!Ztny%e37<6u36W;(nqv%scspjy~ph5|NO)CEAdMg?w
z!y5n2SH&Bsm>uK>om<L2$It4#IY+3B?s<b}q{XRXTU^zL!D~s-V}XOt+=3@J=jepl
zW;4>V`ZL6$Yv-LU<XgKp%2@fH-b?|HIkkV>TmD>pX?+r^z;^v<^|w%K%s_BiPI6&D
z$Q{Y_rNBzz1r0~5m~{6PMPW7r5$qpqp22~?DmSz1SejS(^kR-}1sj(OhD#3(VXJ;s
z_QN-Kg+OK#`uG@|0q*z}u1hr}^yrQd7Rx>Mh$5#GcZF?Va7WYP9=tVNx%4_r!S7l*
zw+<IIzf#Lyp+4Ax_DXQ8g4F&d{*Lu2?{!B9^-ztJePA@;RX47!*;oP0a4AMz{HCfJ
z?U=2}v){fRRo1vVp}f1cyY#JpfBiZ1gQSw&Xwwaw6II{&ZWs`YImTyIqYK^-uXN?K
zt}ULEuzQgnG<Y|0(J8Uyd!nlM-7RWIa_3nxsa|ai)ws+0$+Ho!RS-?s`LX`}u(qd-
zCF;GDIVFYzO5L`Z{-MRpVsdYJcY3u)F2nz9V~K;DJ+RDjP=>uwzZt<{VP{<C<1;#V
zj_+iT*+KP#FL&3hQX~;BcaLmEe?B!48v?Hz=0rB!@$2b!85yPH%#KEb7JOoO!|tzR
z)*z>)X56t?_ObVarEjH6bA4I6dlmf#ozSb6&r3d7OT7(p#|HLgdWNj&mj8GS?;P6b
z-W;+68N?0maa3Xc<CDj$I$QtbZT~3Iq&<qWyzTnNkW6&TRv%M7-tCzVCnPAZHO^^D
zeHoHY%sURd8tiwlyVR%N<6<bVAB^333XARO^$6Zxku=wrFfY7p-qp6=qx@b#-tV%V
z6sksA!DFNt`Ks`4p!w0>&~?L}?gypHf$HE|Qd#|Du5am0<FRP4`>*Naup=Mr%zLig
z;4_gNLkd}EsaBd9!MpQ$P9^fw8%y8w(qE@P_;Q8hk}=Mo_vb&<%NtK&?^6$_$uJjf
zLj^?z?B4uv-}Ua4kYm#DR5=v7Xc8y%RT^R71h@Tb^5arc_<^XxdT-!Sz3Z==$8(6+
zyyy1qRw*(+I4qu0yIkIf-UxpRo|$J#I8K(BaX+oivaf0k=hbX84jkM0$nx5dq88Uf
z3dUv51jakeTvjvJK%wC`df82q(^KX8`Bnua;H3@E?Vh;XhPWT4;1fQq?ZU*4<yoaR
zCLQjUNTH1Fte!#CK>?!+M}O^q8bLl|#K8mDaLoU0JIEZqCv?Sk*JyumX!pna_cp=H
zm81muK5(e-!-;E-$#9$0!=F*gSCv>)VoR-6dK9<5IS%h_E|!CBRs~+_+*%q=Ob=f9
zNHVFLI9v%B=9}1QTdH60?i745^I~bKya%qtpb`t-yMDmI+w2nPxv*W6`F>IA!R^vJ
z!M{2#ttU(z%*vfaM8H9H!>)c>VD&TYg_3$jVSa56^bNd`yjyl*v<Bn2#s;BXV2Qh}
zmz0bP%a#`Hj)O|wx4~PY6L{t?DPJtED-L9eNTiL4q14EF49+ZGdKZ{EjYchfdg@X9
zQS2B&*;*bg*H63QR1)%CV*Lkm$l6#91k=JQH?zt>2CqRI19{4a^Pnw3ZX(3DRK8D+
zxQ1J@Sq&OG1ruTuyp<9<BP!JKc(P%8JgpH16C3~ax_V|YpmCqFONs$K7eD;{ApW!l
zg-u#tL?eb5S%=x42oD!*Q`-2z&b_J^)5$skUg_hRDt4}wj#qhEV|x_0XG*NS8)T2)
zVlSu&fbz?P{=`F%B3#^$H*lG&w)HiZPQx?~T<GE(*zauoi09!a^1&^AuCBWAaJfBT
z`gc<caet+A2*<)INAWW`x|PTrlzy<=6Am|rY=U}FD6Gz<-(#A%S>*j)h!4hFT>tCK
zfhdFm3N8<$M{3&8R!c<1jb8QLr;+(j)7)M%K-_4B`cl4t%>fG-=qTt8AYN}Q<ful{
z(EykyWxmpzb28JQ`w3&jN1*P9Zbi6#a-Nb|AtVLV(0KCag^SZzSo^Hut=$3;kTyyi
zVPb^g2tVHsQ?ISjH^&{}YA8Lo=*Y%vUxk(ALT@ZuPwbiNhd;DBmzH8$NVf+UlR;rf
z+{02~Yb={SkabXUWZ<rI3*K7ls6tb$>KU3)plrz~CMI*Rk*EAMxQ$L4T>PY^jeDS*
zKK=V7I%zpj81tp|w`{%X2@$K?8u3`~WW;bzZH;>O0Hk_L>r=RNrj*GZO$O!VYBZ(R
zbF>P)_xIC!fV2f4fTkULCD0wgz)6Shd@V`hg3xN0iUI8)fPRoffqszTqU!O~)ux+T
rC+7ryzeV4_^#8I6>q|%3F}q_$RX)k#ati2DgWT3M&?r{3fAoI<CwS~V

literal 0
HcmV?d00001

diff --git a/src/statistical-and-mathematical-methods-for-ai/sections/_machine_learning.tex b/src/statistical-and-mathematical-methods-for-ai/sections/_machine_learning.tex
index e1af1db..ccd400d 100644
--- a/src/statistical-and-mathematical-methods-for-ai/sections/_machine_learning.tex
+++ b/src/statistical-and-mathematical-methods-for-ai/sections/_machine_learning.tex
@@ -14,16 +14,20 @@
         $\vec{\uptheta} = \begin{pmatrix} \uptheta_0, \dots, \uptheta_D \end{pmatrix}$ is the parameter vector.
 
     \item[Probabilistic model] \marginnote{Probabilistic model}
-        The model is a multivariate probabilistic distribution.
+        The model is a multivariate probabilistic distribution that 
+        is able to quantify uncertainty in noisy data.
 \end{description}
 
 
 
 \section{Learning}
 
+
 \subsection{Empirical risk minimization}
 \marginnote{Empirical risk minimization}
 Used for function models.
+The parameters of the predictor are directly obtained as an optimization problem that aims to minimize the distance
+between the prediction and the ground truth.
 
 Let $(\vec{x}_n, y_n)$ be a dataset of $N$ elements
 where $\vec{x}_n \in \mathbb{R}^D$ are the examples and $y_n \in \mathbb{R}$ are the labels.
@@ -41,14 +45,14 @@ We denote the output of the estimator as $\hat{y}_n = f_\vec{\uptheta}(\vec{x}_n
         the dataset $(\vec{x}_n, y_n)$ is independent and identically distributed. 
         Therefore, the empirical mean is a good estimate of the population mean.
 
-        \begin{description}
-            \item[Empirical risk] \marginnote{Empirical risk}
-                Given the example matrix $\matr{X} = \begin{pmatrix} \vec{x}_1, \dots, \vec{x}_N \end{pmatrix} \in \mathbb{R}^{N \times D}$
-                and the label vector $\vec{y} = \begin{pmatrix} y_1, \dots, y_N \end{pmatrix} \in \mathbb{R}^N$.
-                The empirical risk is given by the average loss:
-                \[ \textbf{R}_\text{emp}(f_\vec{\uptheta}, \matr{X}, \vec{y}) = \frac{1}{N} \sum_{n=1}^{N} \ell(y_n, \hat{y}_n) \]
+    \item[Empirical risk] \marginnote{Empirical risk}
+        Given the example matrix $\matr{X} = \begin{pmatrix} \vec{x}_1, \dots, \vec{x}_N \end{pmatrix} \in \mathbb{R}^{N \times D}$
+        and the label vector $\vec{y} = \begin{pmatrix} y_1, \dots, y_N \end{pmatrix} \in \mathbb{R}^N$.
+        The empirical risk is given by the average loss:
+        \[ \textbf{R}_\text{emp}(f_\vec{\uptheta}, \matr{X}, \vec{y}) = \frac{1}{N} \sum_{n=1}^{N} \ell(y_n, \hat{y}_n) \]
 
-            \begin{example}[Least-squares loss] \marginnote{Least-squares loss}
+        \begin{description}
+            \item[Least-squares loss] \marginnote{Least-squares loss}
                 The least-squares loss is defined as:
                 \[ \ell(y_n, \hat{y}_n) = (y_n - \hat{y}_n)^2 \]
 
@@ -58,27 +62,207 @@ We denote the output of the estimator as $\hat{y}_n = f_\vec{\uptheta}(\vec{x}_n
                     \min_{\vec{\uptheta} \in \mathbb{R}^D} \frac{1}{N} \sum_{n=1}^{N} (y_n - \vec{\uptheta}^T\vec{x}_n)^2 =
                     \min_{\vec{\uptheta} \in \mathbb{R}^D} \frac{1}{N} \Vert \vec{y} - \matr{X}\vec{\uptheta} \Vert^2
                 \]
-            \end{example}
+        \end{description}
 
-            \item[Expected risk] \marginnote{Expected risk}
-                The expected risk is defined as:
-                \[ \textbf{R}_\text{true}(f_\vec{\uptheta}) = \mathbb{E}_{\vec{x}, y}[\ell(y, f_\vec{\uptheta}(\vec{x}_\text{test}))] \]
-                where the parameters $\vec{\uptheta}$ are fixed and the samples are taken from a test set.
+    \item[Expected risk] \marginnote{Expected risk}
+        The expected risk is defined as:
+        \[ \textbf{R}_\text{true}(f_\vec{\uptheta}) = \mathbb{E}_{\vec{x}, y}[\ell(y, f_\vec{\uptheta}(\vec{x}_\text{test}))] \]
+        where the parameters $\vec{\uptheta}$ are fixed and the samples are taken from a test set.
 
-            \item[Overfitting] \marginnote{Overfitting}
-                A predictor $f_\vec{\uptheta}$ is overfitting when $\textbf{R}_\text{emp}(f, \matr{X}_\text{train}, \vec{y}_\text{train})$
-                underestimates $\textbf{R}_\text{true}(f_\vec{\uptheta})$ (i.e. the loss on the training set is low, but on the test set is high).
+    \item[Overfitting] \marginnote{Overfitting}
+        \sloppy
+        A predictor $f_\vec{\uptheta}$ is overfitting when $\textbf{R}_\text{emp}(f, \matr{X}_\text{train}, \vec{y}_\text{train})$
+        underestimates $\textbf{R}_\text{true}(f_\vec{\uptheta})$ (i.e. the loss on the training set is low, but on the test set is high).
 
-            \item[Regularization] \marginnote{Regularization}
-                Method that introduces a penalty term to the loss that
-                helps to find a compromise between the accuracy and the complexity of the solution:
-                \[ \bar{\ell}(y_n, \hat{y}_n) = \ell(y_n, \hat{y}_n) + \lambda \mathcal{R}(\vec{\uptheta}) \]
-                where $\lambda \in \mathbb{R}^+$ is the regularization parameter and $\mathcal{R}$ is the penalty.
+    \item[Regularization] \marginnote{Regularization}
+        Method that introduces a penalty term to the loss that
+        helps to find a compromise between the accuracy and the complexity of the solution:
+        \[ \bar{\ell}(y_n, \hat{y}_n) = \ell(y_n, \hat{y}_n) + \lambda \mathcal{R}(\vec{\uptheta}) \]
+        where $\lambda \in \mathbb{R}^+$ is the regularization parameter and $\mathcal{R}$ is the regularizer (penalty term).
+
+        \begin{description}
+            \item[Regularized least squares] \marginnote{Regularized least squares} 
+                A simple regularization term for the least squares problem is $\Vert \vec{\uptheta} \Vert^2$.
+                The problem becomes:
+                \[ \min_{\vec{\uptheta} \in \mathbb{R}^D} 
+                    \{ \frac{1}{N} \Vert \vec{y} - \matr{X}\vec{\uptheta} \Vert^2 + \lambda \Vert \vec{\uptheta} \Vert^2 \} \] 
         \end{description}
 \end{description}
 
 
+\subsection{Maximum likelihood estimation (MLE)}
+% \marginnote{Maximum likelihood estimation (MLE)}
+Used for probabilistic models.
+The parameters are determined as the most likely to predict the correct label given an input.
 
-\subsection{Maximum likelihood}
-\marginnote{Maximum likelihood}
-Used for probabilistic models.
\ No newline at end of file
+\begin{description}
+    \item[Negative log-likelihood] \marginnote{Negative log-likelihood}
+        \sloppy
+        Given a random variable $\bm{x}$, a probability density $p_\vec{\uptheta}(\bm{x})$ parametrized by $\vec{\uptheta}$
+        and a predictor, the negative log-likelihood of $\bm{x}$ is:
+        \[ \mathcal{L}_{\bm{x}}(\vec{\uptheta}) = -\log p_\vec{\uptheta}(\bm{x}) \]
+        Note that:
+        \begin{itemize}
+            \item The minus is added as we are converting the problem of maximizing the likelihood to a minimization problem.
+            \item The logarithm is useful for numerical stability.
+        \end{itemize}
+        $\mathcal{L}_{\bm{x}}(\vec{\uptheta})$ indicates how likely it is to observe $\bm{x}$ with
+        $\vec{\uptheta}$ as the parameters of the predictor.
+
+        Given a dataset $(\bm{x}_n, y_n)$ of $N$ independent and identically distributed (i.i.d.) elements,
+        optimizing the likelihood allows to find the most likely parameters to represent the dataset.
+        As the dataset is independent, we have that:
+        \[ p_\vec{\uptheta}(\vec{y} \vert \matr{X}) = \prod_{n=1}^{N} p_\vec{\uptheta}(y_n \vert \bm{x}_n) \]
+        where $\matr{X} = \begin{pmatrix} \bm{x}_1, \dots, \bm{x}_N \end{pmatrix}$ and
+        $\vec{y} = \begin{pmatrix} y_1, \dots, y_N \end{pmatrix}$.
+        Moreover, as the dataset is identically distributed, 
+        each $p_\vec{\uptheta}(y_n \vert \bm{x}_n)$ of the product has the same distribution.
+
+        By applying the logarithm, we have that the negative log-likelihood of a i.i.d. dataset is define as:
+        \[ \mathcal{L}(\vec{\uptheta}) = -\sum_{n=1}^{N} \log p_\vec{\uptheta}(y_n \vert \bm{x}_n) \]
+        and to find good parameters $\vec{\uptheta}$, we solve the problem:
+        \[ 
+            \min_{\vec{\uptheta} \in \mathbb{R}^D} \mathcal{L}(\vec{\uptheta}) =  
+            \min_{\vec{\uptheta} \in \mathbb{R}^D} -\sum_{n=1}^{N} \log p_\vec{\uptheta}(y_n \vert \bm{x}_n) 
+        \]
+
+        \begin{description}
+            \item[Gaussian likelihood] \marginnote{Gaussian likelihood}
+                Using a linear model $\bm{x}^T\vec{\uptheta}$ as predictor and 
+                assuming that the likelihood has a Gaussian distribution as follows:
+                \[ p_\vec{\uptheta}(y_n \,\vert\, \bm{x}_n) = \mathcal{N}(y_n \,\vert\, \bm{x}_n^T\vec{\uptheta}, \sigma^2) \]
+                where the Gaussian distribution has mean $\bm{x}_n^T\vec{\uptheta}$ (i.e. $f_\vec{\uptheta}(\bm{x}_n))$ 
+                and variance $\sigma^2$ for the $n$-th data point.
+        
+                The negative log-likelihood is:
+                \[
+                    \begin{split}
+                        \mathcal{L}(\vec{\uptheta}) &= -\sum_{n=1}^{N} \log p_\vec{\uptheta}(y_n \vert \bm{x}_n) \\
+                            &= -\sum_{n=1}^{N} \log \mathcal{N}(y_n \vert \bm{x}_n^T\vec{\uptheta}, \sigma^2) \\
+                            &= -\sum_{n=1}^{N} \log \left( \frac{1}{\sqrt{2\pi\sigma^2}} \exp\left(-\frac{(y_n-\bm{x}_n^T\vec{\uptheta})^2}{2\sigma^2}\right) \right) \\
+                            &= -\sum_{n=1}^{N} \log\exp\left(-\frac{(y_n-\bm{x}_n^T\vec{\uptheta})^2}{2\sigma^2}\right) - \sum_{n=1}^{N} \log\frac{1}{\sqrt{2\pi\sigma^2}} \\
+                            &= \frac{1}{2\sigma^2} \sum_{n=1}^{N} (y_n-\bm{x}_n^T\vec{\uptheta})^2 - \sum_{n=1}^{N} \log\frac{1}{\sqrt{2\pi\sigma^2}}
+                    \end{split}  
+                \]
+        
+                The minimization problem becomes:
+                \[
+                    \begin{split}
+                        \min_{\vec{\uptheta} \in \mathbb{R}^D} \mathcal{L}(\vec{\uptheta}) &= 
+                            \min_{\vec{\uptheta} \in \mathbb{R}^D} 
+                                \overbrace{\frac{1}{2\sigma^2}}^{\mathclap{\text{constant}}} 
+                                \sum_{n=1}^{N} (y_n-\bm{x}_n^T\vec{\uptheta})^2 - 
+                                \overbrace{\sum_{n=1}^{N} \log\frac{1}{\sqrt{2\pi\sigma^2}}}^{\mathclap{\text{constant}}} \\
+                            &= \min_{\vec{\uptheta} \in \mathbb{R}^D} \sum_{n=1}^{N} (y_n-\bm{x}_n^T\vec{\uptheta})^2 \\
+                            &= \min_{\vec{\uptheta} \in \mathbb{R}^D} \Vert \vec{y} - \matr{\uptheta}\matr{X} \Vert^2
+                    \end{split}    
+                \]
+                which corresponds to the least squares problem.
+        \end{description}
+
+        \begin{figure}[ht]
+            \begin{subfigure}{.45\textwidth}
+                \centering
+                \includegraphics[width=.75\linewidth]{img/gaussian_mle_good.png}
+                \caption{When the parameters are good, the label will be near the mean (i.e. predictor)}
+            \end{subfigure}
+            \hspace*{1em}
+            \begin{subfigure}{.45\textwidth}
+                \centering
+                \includegraphics[width=.75\linewidth]{img/gaussian_mle_bad.png}
+                \caption{When the parameters are bad, the label will be far the mean}
+            \end{subfigure}
+
+            \caption{Geometric interpretation of the Gaussian likelihood. (not sure if this is correct)}
+        \end{figure}
+\end{description}
+
+
+\subsection{Maximum a posteriori estimation (MAP)}
+\marginnote{Maximum a posteriori (MAP)}
+Maximum a posteriori estimation uses the opposite distribution of MLE and maximizes:
+\[ 
+    \max_{\vec{\uptheta} \in \mathbb{R}^D} p(\vec{\uptheta} \vert \matr{X}, \vec{y}) =
+    \min_{\vec{\uptheta} \in \mathbb{R}^D} -p(\vec{\uptheta} \vert \matr{X}, \vec{y})
+\]
+In other words, it maximizes the probability of a set of parameters $\vec{\uptheta}$ given the observation of the dataset $(\matr{X}, \vec{y})$.
+By applying the Bayes' theorem, the problem becomes:
+\[ 
+    \begin{split}
+        \min_{\vec{\uptheta} \in \mathbb{R}^D} 
+            -\frac{p(\vec{y} \vert \matr{X}, \vec{\uptheta}) p(\vec{\uptheta})}{\underbrace{p(\vec{y} \vert \matr{X})}_{\mathclap{\text{constant}}}} &=
+        \min_{\vec{\uptheta} \in \mathbb{R}^D} -p(Y \vert \matr{X}, \vec{\uptheta}) p(\vec{\uptheta}) \\
+        &= \min_{\vec{\uptheta} \in \mathbb{R}^D} \{ -\log p(\vec{y} \vert \matr{X}, \vec{\uptheta}) -\log p(\vec{\uptheta}) \}
+    \end{split}
+\]
+
+\begin{description}
+    \item[Gaussian posteriori] \marginnote{Gaussian posteriori}
+        By assuming that the conditional probability of the dataset follows a Gaussian distribution (as in MLE),
+        the problem becomes:
+        \[ 
+            \min_{\vec{\uptheta} \in \mathbb{R}^D} \{ -\log p(\vec{y} \vert \matr{X}, \vec{\uptheta}) -\log p(\vec{\uptheta}) \} = 
+            \min_{\vec{\uptheta} \in \mathbb{R}^D} \{ \Vert \vec{y} - \matr{\uptheta}\matr{X} \Vert^2 -\log p(\vec{\uptheta}) \} 
+        \]
+
+        Moreover, assuming that $p(\vec{\uptheta}) \sim \mathcal{N}(0, \matr{\Sigma})$, we have that:
+        \[ -\log p(\vec{\uptheta}) = \frac{1}{2\sigma^2} \Vert \vec{\uptheta} \Vert^2 \]
+
+        Therefore, the problem becomes:
+        \[ \min_{\vec{\uptheta} \in \mathbb{R}^D} \{ \Vert \vec{y} - \matr{\uptheta}\matr{X} \Vert^2 + \lambda \Vert \vec{\uptheta} \Vert^2 \} \]
+        MAP can be seen as a regularization factor for MLE.
+\end{description}
+
+
+
+\section{Linear regression}
+\marginnote{Linear regression}
+Given a dataset of inputs $\vec{x}_n \in \mathbb{R}^D$ with corresponding labels $y_n = f(\vec{x}_n) + \varepsilon$,
+where $f: \mathbb{R}^D \rightarrow \mathbb{R}$ and $\varepsilon \sim \mathcal{N}(0, \sigma^2)$ is a Gaussian noise.
+We want to estimate the function $f$.
+
+\begin{description}
+    \item[Model]
+        Because of the noise, we use a probabilistic model with likelihood:
+        \[ p_\vec{\uptheta}(y \,\vert\, \vec{x}) = \mathcal{N}(y \,\vert\, f(\vec{x}), \sigma^2) \]
+        As model, we use a linear predictor:
+        \[ f(\vec{x}) = \vec{x}^T \vec{\uptheta} \]
+
+    \item[Parameter estimation]  
+        To estimate $\vec{\uptheta}$, we can use MLE:
+        \[ \min_{\vec{\uptheta} \in \mathbb{R}^D} -p_\vec{\uptheta}(\vec{y} \vert \matr{X}) \]
+\end{description}
+
+
+\subsection{Maximum likelihood estimation with features}
+\marginnote{MLE with features}
+Linear regression is linear only with respect to the parameters $\vec{\uptheta}$. 
+Therefore, it is possible to apply any transformation to the inputs of $f$ such that:
+\[ f(\vec{x}_n) = (\phi(\vec{x}_n))^T \vec{\uptheta}  \]
+where $\phi: \mathbb{R}^D \rightarrow \mathbb{R}^K$ and $\vec{\uptheta} \in \mathbb{R}^K$.
+
+The likelihood becomes:
+\[ p_\vec{\uptheta}(y \,\vert\, \vec{x}) = \mathcal{N}(y \,\vert\, (\phi(\vec{x}))^T\vec{\uptheta}, \sigma^2) \]
+
+\begin{description}
+    \item[Polynomial regression] \marginnote{Polynomial regression}
+        The transformation function $\phi: \mathbb{R} \rightarrow \mathbb{R}^K$ is defined as:
+        \[  
+            \phi(x) = 
+            \begin{pmatrix}
+                \phi_0(x) \\ \phi_1(x) \\ \phi_2(x) \\ \vdots \\ \phi_{K-1}(x)
+            \end{pmatrix}
+            = 
+            \begin{pmatrix}
+                1 \\ x \\ x^2 \\ \vdots \\ x^{K-1}
+            \end{pmatrix}
+        \]
+        The predictor is then defined as:
+        \[ 
+            \begin{split}
+                f(x) &= (\phi(x))^T \vec{\uptheta} \\
+                    &= \sum_{i=0}^{K-1} \phi_i(x)\vartheta_i \\
+                    &= \sum_{i=0}^{K-1} x^i \vartheta_i
+            \end{split}
+        \]
+\end{description}
\ No newline at end of file