How can multiple distributed entities collaboratively train a shared deep net on their private data while preserving privacy? This paper introduces InstaHide, a simple encryption of training images, which can be plugged into existing distributed deep learning pipelines. The encryption is efficient and applying it during training has minor effect on test accuracy.
InstaHide encrypts each training image with a “one-time secret key” which consists of mixing a number of randomly chosen images and applying a random pixel-wise mask. Other contributions of this paper include: (a) Using a large public dataset (e.g. ImageNet) for mixing during its encryption, which improves security. (b) Experimental results to show effectiveness in preserving privacy against known attacks with only minor effects on accuracy. (c)Theoretical analysis showing that successfully attacking privacy requires attackers to solve a difficult computational problem. (d) Demonstrating that use of the pixel-wise mask is important for security, since Mixupalone is shown to be insecure to some some efficient at-tacks. (e) Release of a challenge dataset1to encourage new attacks. Read More
Monthly Archives: February 2021
Privacy Preserving Machine Learning: Threats and Solutions
For privacy concerns to be addressed adequately in today’s machine learning systems, the knowledge gap between the machine learning and privacy communities must be bridged. This article aims to provide an introduction to the intersection of both fields with special emphasis on the techniques used to protect the data. Read More
Google says it’s too easy for hackers to find new security flaws
Attackers are exploiting the same types of software vulnerabilities over and over again, because companies often miss the forest for the trees.
In December 2018, researchers at Google detected a group of hackers with their sights set on Microsoft’s Internet Explorer. Even though new development was shut down two years earlier it’s such a common browser that if you can find a way to hack it, you’ve got a potential open door to billions of computers.
The hackers were hunting for, and finding, previously-unknown flaws, known as zero-day vulnerabilities.
Google’s security team known as Project Zero, spotlights multiple examples of zero-days, including problems that Google itself has had with its popular Chrome browser. Read More
Project Zero
- Introducing the In-the-Wild Series
- Chrome: Infinity Bug
- Chrome Exploits
- Android Exploits
- Android Post-Exploitation
- Windows Exploits
Tech giants open up about their algorithms
Google, Facebook, TikTok and others are starting to talk more about how their algorithms work in a bid to win trust.
Yes, but: It’s hard to know what isn’t being revealed.
- Google on Monday published a blog post that shows users how to access more information about their search results, the day ahead of its Q4 earnings report.
- Facebook similarly released a post last week about how its News Feed algorithm works the day before its Q4 earnings.
- TikTok last year, amid the threat of a ban from the Trump administration, walked Axios and other reporters through an extensive presentation of how its prized algorithm works.
Reinforcement Learning At Facebook with Jason Gauci
If you ever wanted to learn about machine learning you could do worse than have Jason Gauci teach you. Jason has worked on YouTube recommendations. He was an early contributor to TensorFlow the open-source machine learning platform. His thesis work was cited by DeepMind. Read More
How to learn deep learning by reading papers
Create a system in order to be up to date with deep learning research
Deep learning is moving so fast, that the only way to keep up is by reading directly from the people who publish these new findings. If you’re a technical person and want to learn about deep learning in 2021, you need to read papers. Read More
How AI Accelerates the Fight Against Fake News
Microchips in coronavirus vaccines. Pedophile rings in pizza restaurants. Jewish space lasers. The Internet–bless its heart–has always suffered from its share of wacky wingnuts and conspiracy theories. But in the wake of the November 3 presidential election and the January 6 Capitol riot, social media platforms and governments are stepping up their efforts to crack down on the most problematic content, and AI plays a leading role. Read More
A Day in the Life of Your Data
A Father-Daughter Day at the Playground
Over the past decade, a large and opaque industry has been amassing increasing amounts of personal data. A complex ecosystem of websites, apps, social media companies, data brokers, and ad tech firms track users online and offline, harvesting their personal data. This data is pieced together, shared, aggregated, and monetized, fueling a $227 billion-a-year industry. This occurs every day, as people go about their daily lives, often without their knowledge or permission. Read More
Artificial intelligence and the Gamestonk blowback
Surrounded by rallies of “power to the people,” a rag-tag group of scrappy underdogs recently managed to bring Wall Street to its knees through a dazzling display of disobedient investing that saw Gamestop stocks rocket Moonward. This unprecedented seizure of power by the proletariat has been lauded far and wide as a smack in the mouth for the establishment. Some say it’s a warning shot to the financial kings and queens of the Earth.
… A team of researchers from the University of Gottingen recently converted an algorithmic approach to fighting fake news into a method for detecting online market manipulation. … The relevancy here is that Gamestonk didn’t happen as a result of small-time investment firms fighting against their bigger cousins. Gamestonk was a meme on a message board. Read More
Who gets credit for AI-generated art?
The recent sale of an AI-generated portrait for $432,000 at Christie’s art auction has raised questions about how credit and responsibility should be allocated to individuals involved, and how the anthropomorphic perception of the AI system contributed to the artwork’s success. Here, we identify natural heterogeneity in the extent to which different people perceive AI as anthropomorphic. We find that differences in the perception of AI anthropomorphicity are associated with different allocations of responsibility to the AI system, and credit to different stakeholders involved in art production. We then show that perceptions of AI anthropomorphicity can be manipulated by changing the language used to talk about AI –– as a tool vs agent –– with consequences for artists and AI practitioners. Our findings shed light on what is at stake when we anthropomorphize AI systems, and offers an empirical lens to reason about how to allocate credit and responsibility to human stakeholders. Read More
Rick's Cafe AI 