Join us this week for an in-depth discussion on Compositional Learning in the context of cutting-edge text-to-image generative models. We will explore recent breakthroughs and challenges, focusing on how these models handle compositional tasks and where improvements can be made.
🌀 Abstract: This paper investigates the misuse of text-conditional diffusion models, particularly text-to-image models, which create visually appealing images based on user descriptions. While these images generally represent harmless concepts, they can be manipulated for harmful purposes like propaganda. The authors show that adversaries can introduce biases through backdoor attacks, affecting even well-meaning users. Despite users verifying image-text alignment, the attack remains hidden by preserving the text's semantic content while altering other image features to embed biases, amplifying them by 4-8 times. The study reveals that current generative models make such attacks cost-effective and feasible, with costs ranging from 12 to 18 units. Various triggers, objectives, and biases are evaluated, with discussions on mitigations and future research directions.
Join us this week for an in-depth discussion on Compositional Learning in the context of cutting-edge text-to-image generative models. We will explore recent breakthroughs and challenges, focusing on how these models handle compositional tasks and where improvements can be made.
🌀 Abstract: This paper investigates the misuse of text-conditional diffusion models, particularly text-to-image models, which create visually appealing images based on user descriptions. While these images generally represent harmless concepts, they can be manipulated for harmful purposes like propaganda. The authors show that adversaries can introduce biases through backdoor attacks, affecting even well-meaning users. Despite users verifying image-text alignment, the attack remains hidden by preserving the text's semantic content while altering other image features to embed biases, amplifying them by 4-8 times. The study reveals that current generative models make such attacks cost-effective and feasible, with costs ranging from 12 to 18 units. Various triggers, objectives, and biases are evaluated, with discussions on mitigations and future research directions.
Secure video calling is in high demand. As an alternative to Zoom, many people are using end-to-end encrypted apps such as WhatsApp, FaceTime or Signal to speak to friends and family face-to-face since coronavirus lockdowns started to take place across the world. There’s another option—secure communications app Telegram just added video calling to its feature set, available on both iOS and Android. The new feature is also super secure—like Signal and WhatsApp and unlike Zoom (yet), video calls will be end-to-end encrypted.
Should I buy bitcoin?
“To the extent it is used I fear it’s often for illicit finance. It’s an extremely inefficient way of conducting transactions, and the amount of energy that’s consumed in processing those transactions is staggering,” the former Fed chairwoman said. Yellen’s comments have been cited as a reason for bitcoin’s recent losses. However, Yellen’s assessment of bitcoin as a inefficient medium of exchange is an important point and one that has already been raised in the past by bitcoin bulls. Using a volatile asset in exchange for goods and services makes little sense if the asset can tumble 10% in a day, or surge 80% over the course of a two months as bitcoin has done in 2021, critics argue. To put a finer point on it, over the past 12 months bitcoin has registered 8 corrections, defined as a decline from a recent peak of at least 10% but not more than 20%, and two bear markets, which are defined as falls of 20% or more, according to Dow Jones Market Data.
