条件概率(conditionalprobability)若已知A=ao,则B发生b,的概率称为条件概率,记为(P(b,lao),其中P(b,lao)表示“给定A=ao后B=b的概率”P(ao,b,) = P(b;lao) P (ao) P (ai,b) = P (b;lai) P (ai)→ P(b;) =P(ai, bi) = P(bilai) P(ai) P(ai,b) =P (ailb,) P (b,)b1/4Ab- P(ai) =P(ai,bj) = P(ailb;)P(b))a1/2b1/2b2/3概率树(probabilitytree)1/3bla.11bP(a2) :e.g.P(a1)=P(a3) =2361/b1/311P(b3a1) :P(bila1) =P(b2a1) =ba44"P(a,)1/3b111P(a1,bi)=24-8P(b;la,)
条件概率(conditional probability) 若已知 A=a0,则B 发生 bj 的概率称为条件概率,记为 {P (bj | a0 )} , 其中 P (bj | a0 ) 表示“给定 A=a0 后 B=bj 的概率” 概率树(probability tree) e.g
贝叶斯定理(Baves theorem)P (b;lai) P (ai)P(ailb,) =or P(aib;) α P(b;lai) P(ai)TWOEVENTS:P(6Priorknownknowledge(fixed)THREEEVENTS:P (ai,bi,Ck) = P(ai|bi,Ck) P (bi,Ck) = P (aibi, Ck) P(b,lck) P(ck)= P(ai, bilck)P(ck) = P(ailbi, Ck) P(b;lck)P(ck)P(bi,Chlai) P(ai)P(ai|bj, Ck) = For P(ailbj,ck) α P(bj, chlai)P(ai)P(bi,ck)Priorknownknowledge (fixed)类似可得到多个事件的贝叶斯定理
THREE EVENTS: Prior known knowledge (fixed) 类似可得到多个事件的贝叶斯定理。 贝叶斯定理(Bayes’ theorem) TWO EVENTS: Prior known knowledge (fixed)
例:EachmorningIwalktowork,choosingbetweena longanda shortroute.IfIchoosetheshortroutethenIalways arriveontime,but if ItakethelongroutethenIarrive ontimewithprobability3/4.Thelongrouteisscenic,soIrisktakingitonedayinfour.Ifyouseemearriving ontime,whatprobabilitywouldyouinferformehavingtakenthelongroute?YP (bolat) = 24P(at) =a:远路bo:准时P(bo)7aoA:路线选择B:迟到与否P(bo|as) = 1P(aas:近路bl:迟到33P(as)=l-Pal)P(atlbo)α P(bola)P (az)41615P(bo)=P(bolas)P(as)+P(bola)P(a)316P(asbo)x P(bolas)P(as)4P(ab)元1P(bolat)P(a)P(ailbo)=115P(bo)4天三大P(aslbo)=andP (aalbo)4P(bolas)P(as)5P(aabo):1-P(albo)5P(bo)
例:Each morning I walk to work, choosing between a long and a short route. If I choose the short route then I always arrive on time, but if I take the long route then I arrive on time with probability 3/4 . The long route is scenic, so I risk taking it one day in four. If you see me arriving on time, what probability would you infer for me having taken the long route? al:远路 as:近路 A:路线选择 B:迟到与否 bO:准时 bL:迟到
A.信息的物理本质a.2信息和滴信息是概率的函数,是概率分布对应的摘事件A的所有观测结果为(a)。简单若ao必然发生P(ao)=1→观测给不出信息推理若P(a+ao)很小观测到A≠ao给出大量信息维信息量记为h[P(ai)】口确定A的取值时,若取值概率越小,信息量h越大口对独立事件A和B,若知道A=a,后,进一步知道B=b,值会提供额外的信息(信息量的可加性)(非负)h[P(ai)] = -K log P(ai)h[P(ai,b,)] =h[P(ai) P(b;)] =h[P(ai)] +h[P(b;)]口事件A的信息总量(由A的所有结果的完全集决定)称为信息H(A) =-P(a) logP(a), BitsH(A) =ZP(ai) h[P(ai)He=Hln2000He(A)=-P(ai) lnP(ai), Nats
A.信息的物理本质 a.2 信息和熵 ➢ 信息是概率的函数,是概率分布对应的熵。 事件 A 的所有观测结果为 {ai } 。 若 a0 必然发生 → P(a0 )=1 → 观测给不出信息 若 P(ai ≠a0 )很小 → 观测到 A≠ a0给出大量信息 简单 推理 信息量记为 确定 A 的取值时,若取值概率越小,信息量 h 越大 对独立事件 A 和 B,若知道A=ai 后,进一步知道 B=bj值会提供额外的信息(信息量的可加性) (非负) 事件 A 的信息总量(由 A 的所有结果的完全集决定)称为信息熵