Yarjejeniyoyin Koyon Injin da ba su da Amincewa akan Blockchain na Ethereum

Teburin Abubuwan Ciki

1. Gabatarwa

Wannan bincike ya gabatar da wata sabuwar hanya don ƙirƙirar kwangilomin koyon injin marasa aminci akan blockchain na Ethereum. Tsarin yana ba da damar tantancewa da musayar samfuran koyon inji ta atomatik ta hanyar kwangilomin wayo, wanda ke kawar da haɗarin ɓangarorin biyu kuma yana ƙirƙirar kasuwa mai rarrabawa don hanyoyin warwarewar AI.

2. Bayanan Baya

2.1 Blockchain da Kudi na Sirri

Bitcoin ta gabatar da ajiya da canja wurin kuɗi mai rarrabawa ta amfani da sirrin maɓalli na jama'a da yarda da blockchain. Ethereum ta faɗaɗa wannan iyawa tare da cikakkun kwangilomin wayo na Turing, wanda ke ba da damar rikitattun aikace-aikace masu rarrabawa ciki har da tsarin ajiya da kamfanoni masu rarrabawa.

2.2 Nasarorin Koyon Injin

Nasarar 2012 da Krizhevsky da sauransu suka yi ya nuna cewa GPUs na iya horar da cibiyoyin sadarwar jijiyoyi masu zurfi yadda ya kamata, wanda ya haifar da tsarin AI da ya zarce aikin ɗan adam a wasu ayyuka na musamman kamar rabe-raben hoto, gane murya, da wasan kwaikwayo.

3. Tsarin Fasaha

3.1 Gine-ginen Kwangilar Wayo

Tsarin da aka tsara yana amfani da kwangilomin wayo na Ethereum don ƙirƙirar kasuwa mai rarrabawa inda:

• Masu bayanai za su iya gabatar da ƙalubalen ML tare da lada
• Masu horar da samfura za su iya ƙaddamar da hanyoyin warwarewa
• Tabbatarwar ta atomatik tana tabbatar da daidaiton mafita
• Ana rarraba kuɗaɗen biyan kuɗi ta atomatik

3.2 Tsarin Tabbatar da Samfura

Kwangilar tana amfani da saitin tabbatarwa don tantance samfuran da aka ƙaddamar ta atomatik. Tsarin tabbatarwa yana tabbatar da cewa samfuran suna haɓaka da kyau kuma yana hana wuce gona da iri ta hanyar cikakkun bayanai na gwaji masu zaman kansu.

3.3 Ƙarfafa Tattalin Arziki

Tsarin yana ƙirƙirar farashin da kasuwa ke tuka don albarkatun horarwa na GPU, yana ba masu haƙa ma'adinai damar rarraba kayan aiki bisa ga fa'ida tsakanin haƙa ma'adinan cryptocurrency da horar da koyon injin.

4. Cikakkun Bayanai na Aiwatarwa

4.1 Tushen Lissafi

Ana iya wakiltar tsarin horar da hanyar sadarwar jijiyoyi a matsayin matsalar ingantawa wanda ke rage aikin asara:

$L(\theta) = \frac{1}{m} \sum_{i=1}^{m} L(f(x^{(i)}; \theta), y^{(i)})$

Inda $\theta$ ke wakiltar sigogin samfurin, $m$ shine adadin misalan horo, kuma $L$ shine aikin asara wanda ke kwatanta hasashen $f(x^{(i)}; \theta)$ tare da ainihin alamomin $y^{(i)}$.

4.2 Aiwatar da Lamba

A ƙasa akwai tsarin kwangilar wayo mai sauƙi na Solidity don kasuwar ML:

contract MLMarketplace {
    struct Challenge {
        address owner;
        bytes32 datasetHash;
        uint256 reward;
        uint256 accuracyThreshold;
        bool active;
    }
    
    mapping(uint256 => Challenge) public challenges;
    
    function submitModel(uint256 challengeId, bytes32 modelHash, uint256 accuracy) public {
        require(challenges[challengeId].active, "Challenge not active");
        require(accuracy >= challenges[challengeId].accuracyThreshold, "Accuracy too low");
        
        // Transfer reward to submitter
        payable(msg.sender).transfer(challenges[challengeId].reward);
        challenges[challengeId].active = false;
    }
    
    function createChallenge(bytes32 datasetHash, uint256 accuracyThreshold) public payable {
        uint256 challengeId = nextChallengeId++;
        challenges[challengeId] = Challenge({
            owner: msg.sender,
            datasetHash: datasetHash,
            reward: msg.value,
            accuracyThreshold: accuracyThreshold,
            active: true
        });
    }
}

4.3 Sakamakon Gwaji

An gwada tsarin da aka tsara tare da ayyukan rabe-raben hoto ta amfani da bayanan CIFAR-10. Tabbatarwar da aka yi akan blockchain ta sami daidaiton daidaito da hanyoyin tantancewa na al'ada na tsakiya yayin da ake ba da tabbacin rashin aminci.

5. Bincike da Tattaunawa

Tsarin kwangilar koyon injin maras aminci yana wakiltar babban ci gaba a cikin aikace-aikacen AI masu rarrabawa. Ta hanyar amfani da iyawar kwangilomin wayo na Ethereum, wannan hanya tana magance muhimman batutuwa a cikin haɓaka samfurin ML na al'ada, gami da tabbacin aminci da tabbacin biyan kuɗi. Kamar yadda CycleGAN (Zhu et al., 2017) ta kawo sauyi a fannin fassarar hoto zuwa hoto mara kulawa ta ba da damar horo ba tare da misalan biyu ba, wannan tsarin yana canza haɓaka samfurin ML ta hanyar kawar da buƙatar masu shiga tsakani masu aminci.

Gine-ginen fasaha ya nuna yadda blockchain zai iya samar da sakamakon lissafi da za a iya tantancewa, ra'ayi da ƙungiyoyi kamar Gidauniyar Ethereum suka bincika a cikin bincikensu kan hanyoyin sadarwar oracle masu rarrabawa. Tsarin tattalin arzikin tsarin yana ƙirƙirar tsarin gano farashi na halitta don albarkatun lissafi na GPU, wanda zai iya haifar da mafi ingantaccen rarrabawa tsakanin haƙa ma'adinan cryptocurrency da ayyukan horar da koyon injin. Bisa ga binciken NVIDIA akan lissafin GPU, GPUs na zamani na iya kaiwa har zuwa 125 TFLOPS don ayyukan AI, wanda ya sa su dace da duka algorithms na yarda da blockchain da horar da hanyar sadarwar jijiyoyi.

Idan aka kwatanta da dandamalin ML na al'ada na tsakiya kamar Google's TensorFlow Enterprise ko Amazon SageMaker, wannan hanyar mai rarrabawa tana ba da fa'idodi da yawa: babu batu guda na kasa, bayyananniyar tabbacin samfuri, da samun damar duniya. Duk da haka, ƙalubale sun rage a cikin haɓaka mafita don manyan samfura da cikakkun bayanai saboda farashin gas na Ethereum da iyakokin girman toshe. Ƙirar tsarin ta yi daidai da ƙa'idodin da aka zayyana a cikin takardar farar Ethereum (Buterin, 2014) don ƙirƙirar aikace-aikace masu rarrabawa waɗanda ke aiki ba tare da ɓangarori na uku masu aminci ba.

Tsarin tabbatarwa, duk da yake yana da tasiri ga daidaitattun ayyukan rarrabuwa, na iya buƙatar daidaitawa don ƙarin rikitattun matsalolin ML kamar ƙarfafawa koyo ko hanyoyin sadarwa masu adawa (GANs). Saukoki na gaba na iya haɗa hujjojin sirri marasa sani don tabbatar da samfuri don haɓaka sirri yayin da ake kiyaye tabbatacciya, kama da hanyoyin da ƙungiyoyi kamar Zcash da ƙungiyar Bincike ta Sirri da Haɓaka Ethereum ke haɓakawa.

6. Aikace-aikacen Gaba

Tsarin kwangilar ML maras aminci yana da yuwuwar aikace-aikace masu yawa:

• Kasuwanni na Koyo na Tarayya: Ba da damar horar da samfuri mai kiyaye sirri a cikin tushen bayanai da yawa
• Haɓaka AI ta atomatik: Wakilan software waɗanda ke ƙirƙira da kuma tura samfuran ML ta atomatik
• Hanyoyin Warwarewar ML na Ketare: Haɗin kai tare da sauran hanyoyin sadarwar blockchain don ƙayyadaddun lissafi
• Kasuwanni na Bayanai masu Rarrabawa: Haɗaɗɗun kasuwanni na bayanai da samfura tare da tabbataccen asali
• Haɗin Lissafi na Gefen: Na'urorin IoT waɗanda ke shiga cikin horar da samfurin rarrabawa

7. Bayanan Kara Karatu

1. Buterin, V. (2014). Ethereum: Dandamalin Kwangilar Wayo na Gaba da Dandamalin Aikace-aikace mai Rarrabawa
2. Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Rarraba ImageNet tare da cibiyoyin sadarwar jijiyoyi masu zurfi
3. Zhu, J. Y., Park, T., Isola, P., & Efros, A. A. (2017). Fassarar hoto zuwa hoto mara haɗe ta amfani da hanyoyin sadarwa masu adawa na zagayowar da suka dace
4. Silver, D., et al. (2016). Ƙware wasan Go tare da cibiyoyin sadarwar jijiyoyi masu zurfi da binciken bishiya
5. He, K., Zhang, X., Ren, S., & Sun, J. (2015). Koyo mai saura don gane hoto
6. Hornik, K. (1991). Iyawar kusantar hanyoyin sadarwar ciyarwar da yawa
7. Chung, J. S., Senior, A., Vinyals, O., & Zisserman, A. (2016). Karatun lebe a cikin daji
8. Gidauniyar Ethereum. (2023). Shawarwarin Haɓaka Ethereum
9. Kamfanin NVIDIA. (2023). Lissafin GPU don AI da Koyo mai zurfi