Maswali ya Kujifunza ya Kuimarisha Ajenti nyingi

Majibu na maelezo huonekana unapojibu maswali.

1. Ni nini hufanya mazingira kuwa 'yasiyosimama' kutoka kwa mtazamo wa wakala mmoja katika MARL?
   • Mawakala wengine wanajifunza wakati huo huo na kubadilisha sera zao wenyewe
   • Kitendaji cha zawadi kinawekwa upya bila mpangilio kila kipindi
   • Nafasi ya serikali inakua bila kufungwa kwa muda
   • Sensorer hurejesha uchunguzi wa kelele
2. Je, 'mafunzo ya kati yenye utekelezaji wa madaraka' (CTDE) inamaanisha nini?
   • Mawakala hufunzwa na kukimbia kwenye seva moja kuu
   • Mkosoaji anayeshirikiwa hutumia maelezo ya kimataifa wakati wa mafunzo, lakini mawakala hushughulikia uchunguzi wa ndani wakati wa kusambaza
   • Wakala mkuu mmoja hutoa amri kwa wengine wote wakati wa utekelezaji
   • Data ya mafunzo imewekwa kati lakini zawadi zinagatuliwa
3. Ni mfumo upi wa hisabati unaofanya MDP ya wakala mmoja kuwa mawakala wengi?
   • Mpango wa mstari
   • Mfano wa siri wa Markov
   • Mchezo wa Markov (mchezo wa stochastic)
   • Otomatiki ya hali ya mwisho
4. Katika mpangilio wa MARL wa ushirika, zawadi hupangwaje kwa kawaida?
   • Kila wakala huongeza zawadi ambayo ni hasi ya mwingine
   • Wakala wa haraka pekee ndiye anayepokea zawadi yoyote
   • Zawadi hutolewa kwa nasibu kila hatua
   • Mawakala hushiriki zawadi ya pamoja ya timu wanayoongeza kwa pamoja
5. Which training approach did OpenAI Five use to improve at Dota 2?
   • Self-play against copies of its own current and past policies
   • A fixed lookup table of human moves
   • A supervised-only classifier trained on image labels
   • A single policy trained only against a static scripted opponent
6. Kwa nini 'mgawo wa mkopo' ni mgumu sana katika ushirika wa MARL?
   • Zawadi hufika papo hapo na ni rahisi kufuatilia
   • Ni vigumu kubainisha ni hatua gani za wakala zilichangia matokeo ya pamoja ya timu
   • Kuna wakala mmoja tu anayepokea tuzo
   • Mawakala hawawezi kuona malipo yoyote hata kidogo
7. What structural constraint does QMIX place on its mixing network?
   • It must be a linear sum with fixed weights
   • It ignores all individual agent values
   • The joint action-value is monotonic in each agent’s individual value
   • It requires communication at every action step
8. How does Value Decomposition Networks (VDN) form the team action-value?
   • By multiplying all rewards by the number of agents
   • By selecting the largest local reward only
   • By training a separate team for each agent
   • By summing the individual agents’ action-values