1. Gabatarwa
Saurin karɓar sabis na AI yana canza yanayin trafik a hanyoyin sadarwa gaba ɗaya. Duk da yake a halin yanzu manyan kamfanoni ne ke mamaye sabis na AI, nan gaba yana nuni ga tsarin muhalli mai rarraba inda ƙananan ƙungiyoyi da daidaikun mutane za su iya ɗaukar nauyin nasu samfuran AI. Wannan sauyi yana haifar da manyan ƙalubale wajen daidaita ingancin sabis da jinkiri, musamman a cikin yanayin wayar hannu tare da motsi na masu amfani.
Magungunan da ake da su a cikin ƙididdiga na gefen wayar hannu (MEC) da hanyoyin sadarwa masu cike da bayanai sun gaza saboda ƙayyadaddun zato game da tsarin hanyar sadarwa da motsi na masu amfani. Girman girman samfuran AI na zamani (misali, GPT-4 mai kusan darajar tiriliyan 1.8) ya sa hanyoyin ƙaura na sabis na al'ada suka zama marasa amfani, yana buƙatar ingantattun hanyoyin magance matsalar.
2. Tsarin Matsala
2.1 Tsarin Tsarin
Hanyar sadarwa ta ƙunshi sabar gajimare, tashoshi na tushe, raka'o'in gefen hanya, da masu amfani na wayar hannu tare da zaɓuɓɓukan samfurin AI da aka riga aka horar. Dole ne tsarin ya kula da:
- Yanke shawara game da sanyawa sabis na AI
- Zaɓin sabis ta masu amfani
- Ingantaccen karkatar da buƙatu
- Gudanar da motsi na mai amfani
Muhimman abubuwan da aka haɗa sun haɗa da wuraren ɗaukar hoto na mara waya, hanyoyin haɗin waya tsakanin nodes, da rumbunan adana samfurin AI da aka rarraba.
2.2 Manufar Ingantawa
Tsarin ya tsara matsala mai rikitarwa don daidaita ingancin sabis ($Q$) da jinkiri har zuwa ƙarshe ($L$):
$$\min_{x,y} \alpha \cdot L(x,y) - \beta \cdot Q(x,y) + \gamma \cdot C(x,y)$$
inda $x$ ke wakiltar yanke shawara na sanyawa, $y$ yana nuna ma'auni na karkatacciyar hanya, kuma $C$ yana ɗaukar farashin cunkoso. Matsalar tana la'akari da jinkirin layi mara kyau da ƙayyadaddun iyawa a nodes na hanyar sadarwa.
3. Tsarin da Aka Tsara
3.1 Ramin Trafik don Motsi
Maimakon ƙaura manyan samfuran AI lokacin da masu amfani suka motsa tsakanin wuraren shiga, tsarin yana amfani da ramin trafik. Asalin wurin shiga na mai amfani yana aiki azaman anga, yana karkatar da martani daga sabar nesa zuwa sabon wurin mai amfani. Wannan hanyar tana kawar da ƙaura mai tsada na samfura yayin da take gabatar da ƙarin kayan aikin trafik wanda dole ne a sarrafa su.
3.2 Algorithm na Frank-Wolfe Mai Rarraba
Magani ya samo sharuɗɗan KKT na matakin node kuma ya haɓaka algorithm na Frank-Wolfe mai rarraba tare da sabon ƙa'idar saƙo. Kowane node yana yanke shawara na gida bisa:
$$\nabla f(x^{(k)})^T (x - x^{(k)})$$
inda $f$ shine aikin haƙiƙa kuma $x^{(k)}$ shine maganin na yanzu. Algorithm ɗin yana jujjuya zuwa mafi kyawun na gida yayin kiyaye sarrafa rarraba.
4. Sakamakon Gwaji
Ƙididdiga na lamba sun nuna gagarumin ci gaba fiye da hanyoyin da ake da su:
Rage Jinkiri
Inganci na 35-40% idan aka kwatanta da hanyoyin MEC na al'ada
Ingancin Sabis
Mafi kyawun ma'auni na 15-20% tsakanin daidaito da lokacin amsawa
Sarrafa Motsi
Farashin ƙaura sifili na samfura tare da sarrafa kayan aikin rami
Gwaje-gwajen sun kwatanta hanyoyin sadarwar motoci tare da masu amfani na wayar hannu suna samun damar yin amfani da sabis na AI da yawa. Sakamakon ya nuna tsarin yana sarrafa ma'auni tsakanin ingancin sabis da jinkiri yayin tallafawa motsi na mai amfani.
5. Bincike na Fasaha
Mahimman Bayanai
Mahimmin Bayani: Wannan takarda ta kawo gaskiya mai tsanani—tsarin ƙididdiga na gefen al'ada sun lalace gaba ɗaya don AI mai rarraba. Giwa a cikin daki? Ba za ku iya ƙaura samfuran tiriliyan a cikin lokacin gaskiya ba. Hanyar ramin trafik na marubutan ba kawai wayo ba ne; dabarar da ta dace ce wacce ke bayyana yadda abubuwan more rayuwa na yanzu ba su shirya don juyin juya halin AI ba.
Matsalar Hankali: Hujja tana ci gaba da daidaitaccen tiyata: gano sabani girman girman AI na motsi → ƙi ƙaura a matsayin mara yuwuwa → ba da shawarar rami a matsayin madadin kawai mai yuwuwa → gina tsarin lissafi a kusa da wannan ƙuntatawa. Ba kamar ayyukan ilimi waɗanda suka yi watsi da ƙuntatawa na duniya ba, wannan takarda ta fara ne daga ƙaƙƙarfan iyaka kuma ta yi aiki a baya—daidai yadda ya kamata a yi aikin injiniya.
Ƙarfi & Kurakurai: Aiwar Frank-Wolfe mai rarraba sabon abu ne na gaske, yana gujewa matsalolin tattarawa waɗanda ke addabar yawancin binciken AI na gefe. Duk da haka, hanyar ramin tana jin kamar tura guntun hanya—a ƙarshe, waɗannan ƙarin tafiye-tafiye za su haifar da mafarkin cunkoson su. Takardar ta yarda da haka amma ta raina yadda saurin hanyoyin sadarwa suka ƙaru don ɗaukar yanayin trafik na AI, kamar yadda aka gani a cikin aikin Google na baya-bayan nan kan ƙaddarar rarraba.
Bayanai masu Aiki: Ya kamata masu aikin wayar hannu su gwada wannan hanyar nan da nan don sabis na AI masu sauƙi yayin haɓaka ƙarin mafita na asali don manyan samfura. Ƙa'idar saƙo na iya zama ma'auni don daidaitawar AI mai rarraba, kamar yadda HTTP ya zama don trafik na yanar gizo. Ya kamata masu bincike su mai da hankali kan hanyoyin haɗaka waɗanda ke haɗa rami tare da zaɓin ƙaura na muhimman sassan samfura.
Misalin Tsarin Bincike
Nazarin Shari'a: Hanyar Sadarwar Motar Kai
Yi la'akari da garken motoci masu cin gashin kansu waɗanda ke buƙatar gano abu na ainihi. Ta amfani da tsarin da aka tsara:
- An sanya samfuran AI da yawa (YOLOv7, Detectron2, samfuran al'ada) a ko'ina cikin sabar gefe
- Motoci suna zaɓar samfura bisa ga buƙatun daidaito/jinkiri na yanzu
- Yayin da motoci ke motsawa tsakanin hasumiya na salula, ramin trafik yana kula da haɗin kai zuwa ga asalin masu ɗaukar sabis na AI
- Algorithm mai rarraba yana ci gaba da inganta sanyawa da yanke shawara na karkatacciyar hanya
Wannan hanyar tana guje wa canja wurin samfuran AI na gigabyte da yawa yayin tabbatar da ingancin sabis akai-akai yayin abubuwan motsi.
6. Ayyukan Gaba
Tsarin yana da muhimman tasiri ga fasahohi masu tasowa:
- 6G Hanyoyin Sadarwa: Haɗin kai tare da yankakken hanyar sadarwa don tabbacin sabis na AI
- Ayyukan Metaverse: Sabis na AI masu ƙarancin jinkiri don muhallin nutsewa
- Koyo na Tarayya: Haɗin kai tsakanin horar da samfurin rarraba da ƙaddarwa
- Tsarin IoT: Sabis na AI masu yawa don biliyoyin na'urori masu haɗin kai
- Amsa Gaggawa: Hanyoyin sadarwar AI na gaggawa don yanayin bala'i tare da ƙarancin haɗin kai
Bincike na gaba ya kamata ya magance haɓakawa zuwa hanyoyin sadarwa masu yawa da haɗin kai tare da dabarun matsa lamba na samfurin AI masu tasowa.
7. Nassoshi
- OpenAI. "Rahoton Fasaha na GPT-4" (2023)
- Zhu et al. "Edge AI: Kan-Buƙatar Haɓaka Ƙaddarwar Cikakken Neurci ta hanyar Ƙididdiga na Edge" IEEE Transactions on Wireless Communications (2020)
- Mao et al. "Rarraba Albarkatun don Hanyoyin Sadarwar Ƙididdiga na Edge na Wayar Hannu tare da Girbin Makamashi" IEEE Journal on Selected Areas in Communications (2021)
- Google Research. "Hanyoyi: Rarraba Dataflow na Asynchronous don ML" (2022)
- Ma'auni na IEEE don Ƙididdiga na Edge na Wayar Hannu. "Tsari da Tsarin Tunani" (2023)
- Zhang et al. "CycleGAN: Fassarar Hoto zuwa Hoto mara Biyu ta amfani da Cibiyoyin Adawa masu Juyawa" ICCV (2017)
- 3GPP. "Nazari kan Yanayi da Bukatun don Fasahorin Samun Layi na Gaba" TR 38.913 (2024)