Teburin Abubuwan Ciki
1. Gabatarwa
1.1 Dalilai
Haɗuwar fasahar wucin gadi da blockchain yana ba da dama ta musamman don magance manyan ƙalubale a duka fagagen. Hakar kuɗin crypto, musamman ma hanyoyin Shaidar Aiki (PoW), yana cinye ɗimbin makamashi—amfani da wutar lantarki na Bitcoin a shekara ya wuce na Sweden (131.79 TWh) a cikin 2022. A halin yanzu, horar da AI yana buƙatar albarkatun lissafi masu yawa, tare da farashin horar da ChatGPT ya wuce dala miliyan 5 kuma farashin aiki na yau da kullun ya kai dala 100,000 kafin matakan amfani na yanzu.
1.2 Bayanin Matsala
Manyan ƙalubale guda uku sun haifar da gibin tsakanin AI da hakar kuɗin crypto: (1) rashin ingancin makamashi na yarjejeniyar PoW, (2) rashin amfani da albarkatun lissafi bayan canjin Ethereum zuwa PoS, da (3) manyan shingayen shiga ga ci gaban AI saboda farashin lissafi.
Amfani da Makamashi
131.79 TWh - Amfani da makamashi na Bitcoin na 2022
Hashrate da ba a amfani da shi
1,126,674 GH/s - Ana samun bayan canjin Ethereum zuwa PoS
Farashin Horar da AI
$5M+ - Kudadden horar da ChatGPT
2. Ƙa'idar Shaidar Horarwa
2.1 Ɗabarun Gine-gine
Ƙa'idar PoT tana amfani da hanyar yarjejeniya ta Haƙiƙanin Jurewar Byzantine (PBFT) don daidaita yanayin duniya. Tsarin gine-ginen ya ƙunshi manyan sassa guda uku: rukunin horo rarrabuwa, masu tabbatar da yarjejeniya, da sabar haɗa samfura.
2.2 Aiwar Fasaha
Ƙa'idar tana aiwatar da hanyar sadarwar horo mara tsari (DTN) wanda ke ɗaukar PoT don daidaita horar da samfurin AI rarrabuwa. Tushen ilimin lissafi ya haɗa da haɗakar gradient da hanyoyin tabbatar da samfuri.
Tsarin Lissafi
Haɗakar gradient tana bin dabarar:
$\\theta_{t+1} = \\theta_t - \\eta \\cdot \\frac{1}{N} \\sum_{i=1}^N \\nabla L_i(\\theta_t)$
Inda $\\theta$ ke wakiltar sigogin samfuri, $\\eta$ shine ƙimar koyo, kuma $L_i$ shine aikin asara ga ma'aikaci $i$.
Lambar rubutu: Algorithm na Yarjejeniyar PoT
function PoT_Consensus(training_task, validators):
# Fara horo rarrabuwa
model = initialize_model()
for epoch in range(max_epochs):
# Rarraba samfuri zuwa masu haƙa ma'adinai
gradients = []
for miner in mining_nodes:
gradient = miner.compute_gradient(model, training_task)
gradients.append(gradient)
# Tabbatar da gradients ta amfani da PBFT
if PBFT_validate(gradients, validators):
aggregated_gradient = aggregate_gradients(gradients)
model.update(aggregated_gradient)
# Rarraba lada bisa ga gudunmawa
distribute_rewards(gradients, mining_nodes)
return trained_model
3. Sakamakon Gwaji
3.1 Ma'aunin Aiki
Kimantawar ƙa'idar ta nuna gagarumin ci gaba a cikin aikin aiki, ƙarfin tsari, da tsaron hanyar sadarwa. Hanyar sadarwar horo mara tsari ta sami kashi 85% na aikin madadin tsakiya yayin amfani da kayan aikin hakar ma'adinai da suka dade.
3.2 Kimanta Tsarin
Sakamakon gwaji ya nuna cewa ƙa'idar PoT tana nuna yuwuwar girma dangane da amfani da albarkatu da ingancin farashi. Tsarin ya ci gaba da kasancewa a kan layi na kashi 99.2% yayin gwajin damuwa tare da rukunoni 1,000+ na horo a lokaci guda.
Muhimman Hasashe
- Kashi 85% na aiki idan aka kwatanta da horon tsakiya
- Kashi 99.2% na tsarin a kan layi a ƙarƙashin kaya
- Rage farashin lissafi da kashi 60%
- Tallafawa rukunoni 1,000+ a lokaci guda
4. Bincike na Fasaha
Ƙa'idar Shaidar Horarwa ta wakilci wani sabon ƙirƙira a cikin lissafi rarrabuwa, tana haɗa fagage biyu na fasaha masu saurin ci gaba. Kamar yadda CycleGAN (Zhu et al., 2017) ya nuna fassarar hoto-zuwa-hoto mara kulawa, PoT yana ba da damar sake amfani da kayan aikin lissafi ba tare da buƙatar canje-canje na asali ga kayan aikin da ake da su ba. Amfani da ƙa'idar ta yarjejeniyar PBFT ya yi daidai da kafaffen binciken tsare-tsare rarrabuwa daga ƙungiyoyi kamar Laboratory na Kimiyyar Kwamputa da Wucin Gadi na MIT, wanda ya yi nazari sosai kan jurewar ɓarna a cikin hanyoyin sadarwa rarrabuwa.
Daga mahangar fasaha, PoT yana magance matsalar "aiki mai amfani" wacce ta addabi Tsarin Shaidar Aiki tun daga farkonsu. Ba kamar PoW na al'ada ba inda ƙoƙarin lissafi ke aiki don dalilai na tsaro kawai, PoT yana tura wannan ƙoƙarin zuwa ga horon samfurin AI mai amfani. Wannan hanya tana da kamanceceniya ta falsafa da aikin DAWNBench na Stanford, wanda ya mayar da hankali kan sanya horon koyo mai zurfi ya zama mai sauƙi da inganci, kodayake PoT ya faɗaɗa wannan ra'ayi zuwa kayan aikin da ba su da tsari.
Tasirin tattalin arziki yana da girma. Ta hanyar ƙirƙirar kasuwa don horar da AI rarrabuwa, PoT zai iya ƙarfafa samun damar albarkatun lissafi kamar dandamalin lissafi na girgije (AWS, Google Cloud) amma tare da mulki mara tsari. Duk da haka, ƙalubale sun rage a cikin sirrin samfuri da tabbatarwa—alamomin da masu bincike a cibiyoyi kamar Laboratory na Lissafi Rarrabuwa na EPFL suke magance ta hanyar lissafi mai tsaro na ɗimbin ɓangarori da shaidar rashin sani.
Idan aka kwatanta da hanyoyin koyo na tarayya da Binciken Google ya fara, PoT yana gabatar da ƙwararrun ƙwararrun tushen blockchain waɗanda zasu iya magance matsalar rumbun bayanai yayin tabbatar da biyan diyya ga mahalarta. Nasarar ƙa'idar za ta dogara da cimma daidaito mai sauƙi tsakanin ingancin lissafi, garanti na tsaro, da ƙwararrun tattalin arziki—ƙalubalen da ke kama da matsalolin ingantawa da ake fuskanta a cikin horar da rikitattun hanyoyin sadarwa da kansu.
5. Aikace-aikacen Gaba
Ƙa'idar PoT tana buɗe hanyoyi masu yawa masu ban sha'awa ga ci gaba na gaba:
- Haɗin Kai Tsakanin Silsila: Miƙa PoT zuwa hanyoyin sadarwar blockchain da yawa don ƙirƙirar kasuwar lissafi ɗaya
- Ingantaccen Kayan Aiki na Musamman: Haɓaka ASICs da aka ƙera musamman don horar da AI a cikin tsarin PoT
- Haɓaka Koyo na Tarayya: Haɗa PoT tare da dabarun kiyaye sirri don aikace-aikacen bayanai masu mahimmanci
- Haɗin Kai na Lissafi na Gefe: Tura ƙananan rukunonin PoT akan na'urori na gefe don aikace-aikacen IoT
- Yunƙurin AI mai gina muhalli: Yin amfani da hanyoyin samar da makamashi mai sabuntawa don kayan aikin horar da AI mai dorewa
Waɗannan aikace-aikacen zasu iya yin tasiri sosai a masana'antu ciki har da kiwon lafiya (nazarin hoton likita rarrabuwa), kuɗi (horon samfurin gano zamba), da tsare-tsare masu cin gashin kansu (horon kwaikwayo rarrabuwa).
6. Nassoshi
- Zhu, J. Y., Park, T., Isola, P., & Efros, A. A. (2017). Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks. IEEE International Conference on Computer Vision (ICCV).
- Buterin, V. (2014). A Next-Generation Smart Contract and Decentralized Application Platform. Ethereum White Paper.
- Cambridge Bitcoin Electricity Consumption Index. (2023). University of Cambridge.
- OpenAI. (2023). ChatGPT: Optimizing Language Models for Dialogue.
- Hive Blockchain Technologies. (2023). HPC Strategy Update.
- Lamport, L., Shostak, R., & Pease, M. (1982). The Byzantine Generals Problem. ACM Transactions on Programming Languages and Systems.
- McMahan, B., Moore, E., Ramage, D., Hampson, S., & y Arcas, B. A. (2017). Communication-Efficient Learning of Deep Networks from Decentralized Data. Artificial Intelligence and Statistics.
- Stanford DAWNBench. (2018). An End-to-End Deep Learning Benchmark Suite.
- EPFL Distributed Computing Laboratory. (2022). Secure Multi-Party Computation for Machine Learning.