This media is not supported in your browser
VIEW IN TELEGRAM
Sample game by tensor flow
Interesting 😊😊😊
Interesting 😊😊😊
Forwarded from Ali Fazeli
در تنسورفلو هاب داشتم گشت میزدم به یه ماژول جالبی برخورد کردم. گفتم به اشتراک بگذارم .
https://www.tensorflow.org/tutorials/image_retraining
ماژولی به نام retrain که میتونه برای #یادگیری_انتقالی (استفاده از مدل آموزش داده شده برای ایجاد مدل ای جدید) استفاده بشه و از دیتاست محدودی که بهش داده میشه کلاس جدید استخراج (ایجاد) کنه.
برای اطلاعات بیشتر لینک زیر جالب توجه است:
https://codelabs.developers.google.com/codelabs/tensorflow-for-poets/#0
https://www.tensorflow.org/tutorials/image_retraining
ماژولی به نام retrain که میتونه برای #یادگیری_انتقالی (استفاده از مدل آموزش داده شده برای ایجاد مدل ای جدید) استفاده بشه و از دیتاست محدودی که بهش داده میشه کلاس جدید استخراج (ایجاد) کنه.
برای اطلاعات بیشتر لینک زیر جالب توجه است:
https://codelabs.developers.google.com/codelabs/tensorflow-for-poets/#0
Real_World_Robot_Learning_TensorFlow_Dev_Summit_2018_.136.mp4
27.6 MB
Real-World Robot Learning (TensorFlow Dev Summit 2018)
توسعه دهندگان تنسور فلو
توسعه دهندگان تنسور فلو
Forwarded from Ali Fazeli
#پیشنهاد
مقاله ای میخوندم در مورد امنیت صحت خروجی DNN
بنام One pixel attack for fooling deep neural networks
(البته برای یه ماه و نیم پیشه احتمالا اساتید دیده باشن)
https://arxiv.org/abs/1710.08864
ایده مقاله اینه که با علم به اینکه تغییر دادن یه مقدار خاصی از داده تست (تصویر RGB) میتونه باعث شه که CNN گول بخوره و نتونه درست دسته بندی کنه، چطور میتونیم به حداقل تعداد تغییر در پیکسل برسیم که باعث بشیم CNN اشتباه کنه. اومدن کلاس های احتمالی ای که CNN پیشنهاد داده رو نگاه کردن، با استفاده از یه الگوریتمی بنام دیفرانسیل تکاملی یکسری عکس رندوم (تو داده تست) که هربار یه تغییری رندوم تو تصویر میده رو ایجاد کردن و فید کردن به CNN. اگه صحت رو کاهش داد جایگزین میشه با داده تست قبلی. و این فرایند تکاملی وار ادامه پیدا پیدا میکنه تا صحت (accuracy) کمینه بشه و در عین حال از تغییرات پیکسلی کمتری استفاده بشه.
خود Differential evolution برای من تازگی داشت و دارم در موردش میخونم.
—---------------------------------------------------------------------------------------—
اخیرا پیاده سازی خیلی خوبی برای این مقاله تو گیت هاب دیدم در لینک زیر
https://github.com/Hyperparticle/one-pixel-attack-keras
با استفاده از keras روی cifar10 پیاده سازی شده.
اومده برای معماری های رایج با اسفاده از تغییر دادن یک پیکسل، سه پیکسل و پنج پیکسل حمله رو انجام داده و میزان صحت اون کلاس غلط رو گزارش کرده. تو ریپازیتوری نسخه pretrain شده LeNet و ResNet هست.
برای دانلود بقیه معماری ها (که حجیم ان) از لینک زیر:
https://www.dropbox.com/sh/dvatkpjl0sn79kn/AAC9L4puJ_sdFUkDZfr5SFkLa?dl=0
و قرار دادنشون تو پوشه models
—-------------------------------------------------------------------------------------
کارهای بیشتر
روی حمله هدف گذاری شده اش پیاده سازی کاملی انجام گرفته. توی هر تکرار میاد میزان صحت دسته بندی کلاسی که میخواد به قصد CNN رو مجبور به غلط دسته بندی کردنش داره رو نیگاه میکنه و با DE سعی در کمینه کردن اش داره. اما یه قسمتی هم داره که untargeted attack هست. روی اون میشه کار کرد. یعنی عکس لاک پشت رو دادیم. مهم نیست با چه label ای غلط دسته بندی کنه، فقط میخوایم اشتباه کنه. متد ای داره ژوپیتر نوت بوک اش بنام attck که داخل اش میشه از الگوریتم جدیدی برای هوشمندتر کردنش استفاده کرد.
مقاله ای میخوندم در مورد امنیت صحت خروجی DNN
بنام One pixel attack for fooling deep neural networks
(البته برای یه ماه و نیم پیشه احتمالا اساتید دیده باشن)
https://arxiv.org/abs/1710.08864
ایده مقاله اینه که با علم به اینکه تغییر دادن یه مقدار خاصی از داده تست (تصویر RGB) میتونه باعث شه که CNN گول بخوره و نتونه درست دسته بندی کنه، چطور میتونیم به حداقل تعداد تغییر در پیکسل برسیم که باعث بشیم CNN اشتباه کنه. اومدن کلاس های احتمالی ای که CNN پیشنهاد داده رو نگاه کردن، با استفاده از یه الگوریتمی بنام دیفرانسیل تکاملی یکسری عکس رندوم (تو داده تست) که هربار یه تغییری رندوم تو تصویر میده رو ایجاد کردن و فید کردن به CNN. اگه صحت رو کاهش داد جایگزین میشه با داده تست قبلی. و این فرایند تکاملی وار ادامه پیدا پیدا میکنه تا صحت (accuracy) کمینه بشه و در عین حال از تغییرات پیکسلی کمتری استفاده بشه.
خود Differential evolution برای من تازگی داشت و دارم در موردش میخونم.
—---------------------------------------------------------------------------------------—
اخیرا پیاده سازی خیلی خوبی برای این مقاله تو گیت هاب دیدم در لینک زیر
https://github.com/Hyperparticle/one-pixel-attack-keras
با استفاده از keras روی cifar10 پیاده سازی شده.
اومده برای معماری های رایج با اسفاده از تغییر دادن یک پیکسل، سه پیکسل و پنج پیکسل حمله رو انجام داده و میزان صحت اون کلاس غلط رو گزارش کرده. تو ریپازیتوری نسخه pretrain شده LeNet و ResNet هست.
برای دانلود بقیه معماری ها (که حجیم ان) از لینک زیر:
https://www.dropbox.com/sh/dvatkpjl0sn79kn/AAC9L4puJ_sdFUkDZfr5SFkLa?dl=0
و قرار دادنشون تو پوشه models
—-------------------------------------------------------------------------------------
کارهای بیشتر
روی حمله هدف گذاری شده اش پیاده سازی کاملی انجام گرفته. توی هر تکرار میاد میزان صحت دسته بندی کلاسی که میخواد به قصد CNN رو مجبور به غلط دسته بندی کردنش داره رو نیگاه میکنه و با DE سعی در کمینه کردن اش داره. اما یه قسمتی هم داره که untargeted attack هست. روی اون میشه کار کرد. یعنی عکس لاک پشت رو دادیم. مهم نیست با چه label ای غلط دسته بندی کنه، فقط میخوایم اشتباه کنه. متد ای داره ژوپیتر نوت بوک اش بنام attck که داخل اش میشه از الگوریتم جدیدی برای هوشمندتر کردنش استفاده کرد.
GitHub
GitHub - Hyperparticle/one-pixel-attack-keras: Keras implementation of "One pixel attack for fooling deep neural networks" using…
Keras implementation of "One pixel attack for fooling deep neural networks" using differential evolution on Cifar10 and ImageNet - Hyperparticle/one-pixel-attack-keras
#خبر و چالش
توجه ویژه مسترکارت به سیستم سکوریتی و پسورد کارت های مبتنی بر بیومتریک و هوش مصنوعی
In financial services today, security and innovation can work with each other, and against each other.
FinServ cybersecurity is, of course, a prime target for innovation. But customers’ constant demand for cutting-edge products and services is adding to the load of already heavy security burdens. At the same time, providers must ensure a positive, consistent customer experience.
Mastercard Chief Security Solutions Officer Ajay Bhalla recently explored the relationship between financial services innovation and security, telling PYMNTS how these two dynamics can complement each other while guiding the FinServ space in its journey to a better service offering.
Bank customers’ needs and demands of financial service providers are growing more sophisticated and complex. What is driving this trend?
According to the Digital Evolution Index, more countries in more parts of the world have robust digital ecosystems, driving demand for trusted technology. We are on track toward a world with more than 30 billion connected devices – an irreversible trend that is dramatically changing the digital economy and delivering fulfillment of the possibility that consumers have been waiting for: The ability to increase transparency, interact with brands and move faster, with less friction.
For the banks and merchants serving them, the challenge is matching the great strides in connectivity with consistency of experience and security. The pace of innovation focused on consumer experience is showing no signs of slowing down. Meanwhile, criminals are increasingly using technology designed for good in a nefarious manner. Key to success in 2018 will be the adoption of a security by design approach in innovation that marries the establishment and deployment of standards with leading edge technology.
With so many options available to financial service providers to implement and offer new technologies, how can they provide a consistent end-user experience while ensuring that products and services are cutting-edge?
We’re off to a great start in 2018, securing innovation through globally adopted standards like 3-D Secure 2.0 from EMVCo and “PIN on Glass” from the PCI Council. These create consistency and pave the way for banks and merchants to plug in and leverage the power of billions of connected devices to deliver greater consumer experiences, while also addressing threats from cyberattacks and hacks. Put simply: Industry standards enable trust and make it feasible to scale the use of technology.
Passwords remain commonplace, despite the evolution of security technology. How can financial service providers address customer demand for authentication processes that are more secure, efficient and convenient than passwords?
For consumers – some of which have up to 90 online accounts and find passwords difficult to remember – we have room to improve. For the past five years, we’ve been very focused on getting rid of passwords. Passwords are “something you know,” not “something you are.” Remembering passwords is frustrating, and only getting harder as digital activities proliferate. Identifying people by who they are makes life safer and simpler for the individual.
That is the driving force behind our efforts to develop Identity Check, which uses biometric identifiers, such as fingerprint, iris, facial or voice, to verify consumers’ identities using a mobile device during online shopping and banking activities. This speeds up checkout time, improves security and reduces cart abandonment rates. Our biometric payment card, which has already been trialed with consumers, is another example of reshaping their experience. And by retaining focus on consumer-centric solutions, we are not restricted to the single biometric of any device.
These technologies are on the edge of innovation, and research we conducted in partnership with the University of Oxford found that 92 percent of banking professionals want to adopt biometric technology. 2
توجه ویژه مسترکارت به سیستم سکوریتی و پسورد کارت های مبتنی بر بیومتریک و هوش مصنوعی
In financial services today, security and innovation can work with each other, and against each other.
FinServ cybersecurity is, of course, a prime target for innovation. But customers’ constant demand for cutting-edge products and services is adding to the load of already heavy security burdens. At the same time, providers must ensure a positive, consistent customer experience.
Mastercard Chief Security Solutions Officer Ajay Bhalla recently explored the relationship between financial services innovation and security, telling PYMNTS how these two dynamics can complement each other while guiding the FinServ space in its journey to a better service offering.
Bank customers’ needs and demands of financial service providers are growing more sophisticated and complex. What is driving this trend?
According to the Digital Evolution Index, more countries in more parts of the world have robust digital ecosystems, driving demand for trusted technology. We are on track toward a world with more than 30 billion connected devices – an irreversible trend that is dramatically changing the digital economy and delivering fulfillment of the possibility that consumers have been waiting for: The ability to increase transparency, interact with brands and move faster, with less friction.
For the banks and merchants serving them, the challenge is matching the great strides in connectivity with consistency of experience and security. The pace of innovation focused on consumer experience is showing no signs of slowing down. Meanwhile, criminals are increasingly using technology designed for good in a nefarious manner. Key to success in 2018 will be the adoption of a security by design approach in innovation that marries the establishment and deployment of standards with leading edge technology.
With so many options available to financial service providers to implement and offer new technologies, how can they provide a consistent end-user experience while ensuring that products and services are cutting-edge?
We’re off to a great start in 2018, securing innovation through globally adopted standards like 3-D Secure 2.0 from EMVCo and “PIN on Glass” from the PCI Council. These create consistency and pave the way for banks and merchants to plug in and leverage the power of billions of connected devices to deliver greater consumer experiences, while also addressing threats from cyberattacks and hacks. Put simply: Industry standards enable trust and make it feasible to scale the use of technology.
Passwords remain commonplace, despite the evolution of security technology. How can financial service providers address customer demand for authentication processes that are more secure, efficient and convenient than passwords?
For consumers – some of which have up to 90 online accounts and find passwords difficult to remember – we have room to improve. For the past five years, we’ve been very focused on getting rid of passwords. Passwords are “something you know,” not “something you are.” Remembering passwords is frustrating, and only getting harder as digital activities proliferate. Identifying people by who they are makes life safer and simpler for the individual.
That is the driving force behind our efforts to develop Identity Check, which uses biometric identifiers, such as fingerprint, iris, facial or voice, to verify consumers’ identities using a mobile device during online shopping and banking activities. This speeds up checkout time, improves security and reduces cart abandonment rates. Our biometric payment card, which has already been trialed with consumers, is another example of reshaping their experience. And by retaining focus on consumer-centric solutions, we are not restricted to the single biometric of any device.
These technologies are on the edge of innovation, and research we conducted in partnership with the University of Oxford found that 92 percent of banking professionals want to adopt biometric technology. 2
018 is the year we can scale the adoption of biometrics in payments.
Artificial intelligence is another technology that’s driving advancement of financial service cybersecurity. How will FinServ players deploy AI tools in the coming year?
Artificial intelligence and machine learning are increasingly providing the means to interpret and action the complexity and scale of data to generate useful intelligence and an individualized outcome for every interaction, despite the scale/volume. For us, artificial intelligence is already embedded in our network and how we help protect our partners. The acquisition of Brighterion progressed our use of AI, enabling our network to adapt faster and deliver even greater accuracy in decisioning and managing risk.
AI is increasingly being deployed to help banks and merchants look at these new data points to give people a better experience. This will continue to grow in 2018, but the shift will be focused on leveraging the Internet of Things through behavioral analytics. Our acquisition of NuData Security has improved visibility into what is happening at the device level. It allows us to assess patterns of behavior of the user, detect botnet automation and determine the reputation of the device – critical insight for merchants in optimizing the consumer experience and for issuers in determining whether to approve payment. This capability can be applied across any device in the future.
What barriers will the financial services space have to overcome as it integrates and experiments with new technologies to enhance both the customer experience and security?
The internet holds the promise of a bright future ahead, but between here and there is a chasm – and it’s called insecurity. The investment in technology, such as physical biometrics, advanced behavioral analytics, risk-based authentication and artificial intelligence, will grow. It has to, in order to secure digital payments, mobile banking applications and other channels across established, challenger and FinTech sectors. Securing innovation and innovating security will help get us across the chasm and deliver greater trust in the right technologies in 2018.
منبع:
https://www.pymnts.com/news/b2b-payments/2018/mastercard-finserv-innovation-security-ai-b2b/
Artificial intelligence is another technology that’s driving advancement of financial service cybersecurity. How will FinServ players deploy AI tools in the coming year?
Artificial intelligence and machine learning are increasingly providing the means to interpret and action the complexity and scale of data to generate useful intelligence and an individualized outcome for every interaction, despite the scale/volume. For us, artificial intelligence is already embedded in our network and how we help protect our partners. The acquisition of Brighterion progressed our use of AI, enabling our network to adapt faster and deliver even greater accuracy in decisioning and managing risk.
AI is increasingly being deployed to help banks and merchants look at these new data points to give people a better experience. This will continue to grow in 2018, but the shift will be focused on leveraging the Internet of Things through behavioral analytics. Our acquisition of NuData Security has improved visibility into what is happening at the device level. It allows us to assess patterns of behavior of the user, detect botnet automation and determine the reputation of the device – critical insight for merchants in optimizing the consumer experience and for issuers in determining whether to approve payment. This capability can be applied across any device in the future.
What barriers will the financial services space have to overcome as it integrates and experiments with new technologies to enhance both the customer experience and security?
The internet holds the promise of a bright future ahead, but between here and there is a chasm – and it’s called insecurity. The investment in technology, such as physical biometrics, advanced behavioral analytics, risk-based authentication and artificial intelligence, will grow. It has to, in order to secure digital payments, mobile banking applications and other channels across established, challenger and FinTech sectors. Securing innovation and innovating security will help get us across the chasm and deliver greater trust in the right technologies in 2018.
منبع:
https://www.pymnts.com/news/b2b-payments/2018/mastercard-finserv-innovation-security-ai-b2b/
PYMNTS.com
Mastercard Talks Security and Innovation | PYMNTS.com
Credit card company Mastercard's Ajay Bhalla explores the relationship between innovation and security, as customers demand newer tools and as FinServ providers work toward a better, more consistent user experience.
20 مقاله ی برتر یادگیری عمیق
Top 20 #DeepLearning Papers, 2018 Edition http://goo.gl/ScbfwP
از طرف خانم دکتر مهرنوش
Top 20 #DeepLearning Papers, 2018 Edition http://goo.gl/ScbfwP
از طرف خانم دکتر مهرنوش
Forwarded from Ali Fazeli
#پیشنهاد
قابل توجه علاقه مندان به NLP
لینک زیر مجموعه گزارش های دانشجویان دانشگاه استنفورد در مورد دوره زیر است:
Natural Language Processing with Deep Learning
با محوریت کار بر روی دیتاست پرسش و پاسخ کوتاه خود استنفورد (SQuAD)
گزارش کار آقای شریف زاده رو هم بخونین. مقدمه اش جالبه😂
http://web.stanford.edu/class/cs224n/reports.html
قابل توجه علاقه مندان به NLP
لینک زیر مجموعه گزارش های دانشجویان دانشگاه استنفورد در مورد دوره زیر است:
Natural Language Processing with Deep Learning
با محوریت کار بر روی دیتاست پرسش و پاسخ کوتاه خود استنفورد (SQuAD)
گزارش کار آقای شریف زاده رو هم بخونین. مقدمه اش جالبه😂
http://web.stanford.edu/class/cs224n/reports.html
برای دوستانی که به برنامه نویسی با موبایل علاقه مند هستند تنسورفلو یاری ردیف کرده
https://github.com/benoitsteiner/tensorflow-opencl
برای گوشی های اندرویدی که جی پی یو خوبی دارن میتونه جالب باشه👆
با تشکر از دوست عزیز
@AlcatrazJG
https://github.com/benoitsteiner/tensorflow-opencl
برای گوشی های اندرویدی که جی پی یو خوبی دارن میتونه جالب باشه👆
با تشکر از دوست عزیز
@AlcatrazJG
GitHub
GitHub - benoitsteiner/tensorflow-opencl: OpenCL support for TensorFlow
OpenCL support for TensorFlow. Contribute to benoitsteiner/tensorflow-opencl development by creating an account on GitHub.
وبینار آینده بیومتریک روی موبایل
WEBINAR RECORDING: New Frontiers in Mobile Biometrics
Posted on March 29, 2018
Hot on the heels of Mobile World Congress 2018 in Barcelona, FindBiometrics and GSMA hosted the mobile biometrics webinar event of the year: New Frontiers in Mobile Biometrics. The 45 minute webcast began with a review of the biggest identity tech news out of MWC and a presentation from GSMA’s David Pollington highlighting the state of global mobile identity initiatives. The conversation then opened to include industry thought leaders Frances Zelazny, VP, BioCatch, and Fredrik Clementson, R&D Director, Precise Biometrics, for an expert panel discussion that provided insight into key biometrics topics such as:
– How new regulations like PSD2 are affecting the mobile identity management industry
– What’s next for mobile biometrics now that strong authentication is ubiquitous in the smartphone market
– How new innovations in smart cards, AI and connectivity are making biometrics essential to the mobile-first world of consumer finance
– What role standards and specifications bodies play in the new mobile biometrics landscape
– How world leading biometrics vendors are addressing emerging challenges in this dynamic market
در ادامه فیلم مربوط به این وبینار گذاشته خواهد شد.
منبع:
https://findbiometrics.com/webinar-new-frontiers-mobile-biometrics-503299/
WEBINAR RECORDING: New Frontiers in Mobile Biometrics
Posted on March 29, 2018
Hot on the heels of Mobile World Congress 2018 in Barcelona, FindBiometrics and GSMA hosted the mobile biometrics webinar event of the year: New Frontiers in Mobile Biometrics. The 45 minute webcast began with a review of the biggest identity tech news out of MWC and a presentation from GSMA’s David Pollington highlighting the state of global mobile identity initiatives. The conversation then opened to include industry thought leaders Frances Zelazny, VP, BioCatch, and Fredrik Clementson, R&D Director, Precise Biometrics, for an expert panel discussion that provided insight into key biometrics topics such as:
– How new regulations like PSD2 are affecting the mobile identity management industry
– What’s next for mobile biometrics now that strong authentication is ubiquitous in the smartphone market
– How new innovations in smart cards, AI and connectivity are making biometrics essential to the mobile-first world of consumer finance
– What role standards and specifications bodies play in the new mobile biometrics landscape
– How world leading biometrics vendors are addressing emerging challenges in this dynamic market
در ادامه فیلم مربوط به این وبینار گذاشته خواهد شد.
منبع:
https://findbiometrics.com/webinar-new-frontiers-mobile-biometrics-503299/
FindBiometrics
WEBINAR RECORDING: New Frontiers in Mobile Biometrics
Hot on the heels of Mobile World Congress 2018 in Barcelona, FindBiometrics and GSMA hosted the mobile biometrics webinar event of the year: New Frontiers in Mobile Biometrics.
دوستان گلم سلام و عرض ادب یکی از بچه های خوب گروهی، که ایشون در امریکا به سر می برن این متن رو گذاشتن و با اجازه ایشون در اینجا به اشتراک می گذارم....
"
سلام دوستان. بچه های امنیتی نگاهی به برنامه کاراموزی لب ما بندازید. دانجشو میگیریم. اطلاعات کامل.
http://sefcom.asu.edu/internship.html
فقط فرمودند که ای دی اشان رو اشتراک نگذارم و فقط جهت اگاهی شما اعلام نموندند..... امیدوارم شما دوستان عزیز یکی از کسانی باشید که موفق به اپلای در این زمینه باشید. با تشکر از محسن عزیز.
"
سلام دوستان. بچه های امنیتی نگاهی به برنامه کاراموزی لب ما بندازید. دانجشو میگیریم. اطلاعات کامل.
http://sefcom.asu.edu/internship.html
فقط فرمودند که ای دی اشان رو اشتراک نگذارم و فقط جهت اگاهی شما اعلام نموندند..... امیدوارم شما دوستان عزیز یکی از کسانی باشید که موفق به اپلای در این زمینه باشید. با تشکر از محسن عزیز.
Machine learning application (Kartal)
دوستان گلم سلام و عرض ادب یکی از بچه های خوب گروهی، که ایشون در امریکا به سر می برن این متن رو گذاشتن و با اجازه ایشون در اینجا به اشتراک می گذارم.... " سلام دوستان. بچه های امنیتی نگاهی به برنامه کاراموزی لب ما بندازید. دانجشو میگیریم. اطلاعات کامل. ht…
اگر کسی از دوستان تونستن پذیرش اولیه بگیرن حتما بنده رو در جریان بگذارید که شما رو بتونم با آقا محسن عزیز در اون تایم لینک کنم که کمک بیشتر برایتان شود. 🙏🙏🙏