Teburin Abubuwan Ciki
- 1. Gabatarwa
- 2. Tarihi da Matsala
- 3. Hanyar Aiki
- 4. Cikakkun Bayanai na Fasaha da Tsarin Lissafi
- 5. Sakamakon Gwaji
- 6. Misalin Tsarin Bincike
- 7. Muhimman Bayanai
- 8. Bincike na Asali
- 9. Aikace-aikace na Gaba da Hangen Nesa
- 10. Manazarta
1. Gabatarwa
Tsaron kalmar sirri ya kasance babban ƙalubale a tsarin tantancewa. Wannan takarda tana magance sonƙararwa na asali a ƙirƙirar ƙarfin kalmar sirri ta duniya, musamman a hare-haren ƙamus. Marubutan suna ba da shawarar wata sabuwar hanya ta amfani da deep learning da dynamic dictionaries don ƙarin daidaitaccen kwaikwayon dabarun hasashen abokan gaba, rage ƙarin ƙima na ƙarfin kalmar sirri.
2. Tarihi da Matsala
Hare-haren ƙamus na gargajiya sun dogara da tsayayyun tsare-tsare (ƙamus da dokokin canzawa) waɗanda suka kasa kama dabarun daidaitawa na maharan na duniya. Wannan yana haifar da sonƙararwa mai mahimmanci a ƙididdige ƙarfin kalmar sirri, galibi yana ƙara ƙima ga tsaro. Takarda ta nuna cewa masana suna daidaita tsare-tsaren su da hannu, wani tsari mai wuyar kwaikwaya a wuraren bincike.
3. Hanyar Aiki
Hanyar da aka ba da shawara tana sarrafa ƙirƙirar tsare-tsaren hari masu inganci ta amfani da deep learning da daidaitawa ta atomatik.
3.1 Deep Neural Network don Ƙirƙirar Hari
An horar da wata hanyar sadarwa mai zurfi (deep neural network) don ƙirƙirar ƙwarewar abokan gaba wajen gina tsare-tsaren hari. Hanyar sadarwa tana koyon hasashen haɗe-haɗe masu inganci na ƙamus da dokoki bisa ga halayen manufa, tana kawar da buƙatar daidaitawa da hannu.
3.2 Dabarun Hasashen Canzawa
An gabatar da dabarun hasashen canzawa a cikin hare-haren ƙamus. Waɗannan dabarun suna kwaikwayon ikon masana na daidaitawa a kan gudu ta hanyar haɗa ilimi game da manufa (misali, tsarin kalmar sirri da aka fallasa) yayin harin, suna inganta ingancin hasashe.
4. Cikakkun Bayanai na Fasaha da Tsarin Lissafi
Jigon hanyar ya ƙunshi wani tsari mai yiwuwa don hasashen kalmar sirri. Bari $P$ ya zama saitin duk kalmomin sirri masu yiwuwa. Yiwuwar hasashen kalmar sirri $p$ an ƙirƙira shi kamar haka:
$\Pr[\text{hasashe } p] = \sum_{d \in D} \sum_{r \in R} \Pr[d] \cdot \Pr[r|d] \cdot \mathbb{I}[r(d) = p]$
inda $D$ shine saitin ƙamus, $R$ shine saitin dokokin canzawa, kuma $\mathbb{I}$ aiki ne mai nuna alama. Hanyar sadarwa mai zurfi (deep neural network) tana koyon yiwuwar sharadi $\Pr[r|d]$ da $\Pr[d]$ daga bayanan hari na tarihi, tana ba da damar samar da jerin hasashe mafi kyau.
5. Sakamakon Gwaji
An gudanar da gwaje-gwaje akan manyan bayanan kalmar sirri (misali, RockYou, LinkedIn). Hanyar da aka ba da shawara ta ci gaba da yin nasara akan hare-haren ƙamus na tsaye, tana samun har zuwa 30% mafi girman ƙimar nasarar hasashe a daidai adadin hasashe. Hoto na 1 (na ra'ayi) yana nuna kwatancen lanƙwasa hasashe: hanyar canzawa (layi mai ƙarfi) tana riƙe da gangara mai tsayi idan aka kwatanta da tushen tsaye (layi mai ɗigo), yana nuna mafi ingancin dawo da kalmar sirri.
6. Misalin Tsarin Bincike
Nazarin Shari'a: Kwaikwayon Hari Mai Daidaitawa
Yi la'akari da wani tsarin manufa mai jerin kalmomin sirri da aka fallasa na 10,000. Hari na ƙamus na tsaye yana amfani da tsayayyen tsarin dokoki (misali, ƙara lambobi). Hanyar canzawa ta farko tana nazarin jerin da aka fallasa don gano tsarin gama gari (misali, "password123"), sannan ta daidaita tsarin dokoki a hankali don ba da fifiko ga irin waɗannan canje-canje. Wannan yana haifar da ƙarin 40% a kalmomin sirri da aka fasa a cikin hasashe miliyan 1 na farko idan aka kwatanta da tushen tsaye.
7. Muhimman Bayanai
- Rage Sonƙararwa: Hanyar tana rage ƙarin ƙima na ƙarfin kalmar sirri sosai, tana ba da ƙididdige tsaro mafi gaskiya.
- Sarrafa Aiki ta atomatik: Tana kawar da buƙatar daidaitawa da hannu na masana, tana sa hare-haren ci gaba su zama masu isa ga masu bincike.
- Daidaitawa: Dabarun canzawa suna kwaikwayon maharan na duniya sosai, suna inganta ingancin binciken tsaro.
8. Bincike na Asali
Mahimman Bayani: Wannan takarda tana magance wani aibi na asali a binciken tsaron kalmar sirri: dogaro da tsayayyun samfuran hari da masana suka daidaita waɗanda suka kasa wakiltar barazanar duniya. Amfani da marubutan na deep learning don sarrafa da daidaita tsare-tsaren hari ta atomatik babban mataki ne zuwa ga ingancin muhalli a kimantawar tsaro.
Tsarin Tunani: Takarda tana ci gaba a hankali daga gano matsalar sonƙararwa, zuwa ba da shawarar mafita ta koyo, zuwa tabbatar da ita ta zahiri. Haɗin dynamic dictionaries musamman yana da kyau, domin yana kwaikwayon yadda maharan ke tace dabarun su a hankali bisa ga nasarorin da suka samu.
Ƙarfi da Rashi: Babban ƙarfi shine tabbatarwa ta zahiri akan manyan bayanai, yana nuna ingantattun ci gaba. Duk da haka, dogaron hanyar akan bayanan hari na tarihi na iya iyakance ikon ta na gaba ga sabbin tsarin hari. Bugu da ƙari, nauyin lissafi na horar da hanyar sadarwa mai zurfi na iya zama cikas ga wasu masu aiki.
Bayanan Aiki: Ga masu aikin tsaro, wannan aikin yana ba da tsari don mafi daidaitattun ma'aunin ƙarfin kalmar sirri. Ga masu bincike, yana nuna buƙatar ƙaura daga ma'aunin tsaye. Ana iya haɗa hanyar cikin kayan aikin binciken kalmar sirri don ba da ƙididdige haɗari mafi gaskiya. Kamar yadda aka ambata a cikin ayyukan da ke da alaƙa (misali, Goodfellow et al., "Generative Adversarial Nets", NeurIPS 2014), horarwa ta adawa na iya ƙara ƙarfafa samfurin.
9. Aikace-aikace na Gaba da Hangen Nesa
Tsarin da aka ba da shawara yana da fa'idodi masu yawa. Aikin gaba zai iya faɗaɗa ra'ayin dynamic dictionary zuwa wasu fannoni kamar tsaron CAPTCHA ko gano kutse. Haɗin reinforcement learning zai iya ba da damar cikakkun wakilan hari masu cin gashin kansu waɗanda ke koyon mafi kyawun dabarun a ainihin lokaci. Bugu da ƙari, ana iya amfani da hanyar don tsara manufofin kalmar sirri masu ƙarfi ta hanyar gano tsarin da suka fi fuskantar hare-haren daidaitawa.
10. Manazarta
- Pasquini, D., Cianfriglia, M., Ateniese, G., & Bernaschi, M. (2021). Reducing Bias in Modeling Real-world Password Strength via Deep Learning and Dynamic Dictionaries. USENIX Security Symposium.
- Goodfellow, I., Pouget-Abadie, J., Mirza, M., et al. (2014). Generative Adversarial Nets. NeurIPS.
- Ur, B., Segreti, S. M., Bauer, L., et al. (2015). Measuring Real-World Accuracies and Biases in Modeling Password Guessability. USENIX Security Symposium.
- Weir, M., Aggarwal, S., de Medeiros, B., & Glodek, B. (2009). Password Cracking Using Probabilistic Context-Free Grammars. IEEE S&P.