Abstract: Contrastive vision-language representation learning has achieved state-of-the-art performance for zero-shot classification, by learning from millions of image-caption pairs crawled from the internet. However, the massive data that powers large multimodal models such as CLIP, makes them extremely vulnerable to various types of adversarial attacks, including targeted and backdoor data poisoning attacks. Despite this vulnerability, robust contrastive vision-language pretraining against adversarial attacks has remained unaddressed. In this work, we propose RoCLIP, the first effective method for robust pretraining {and fine-tuning} multimodal vision-language models. RoCLIP effectively breaks the association between poisoned image-caption pairs by considering a pool of random examples, and (1) matching every image with the text that is most similar to its caption in the pool, and (2) matching every caption with the image that is most similar to its image in the pool.

But the constant and surprising raft of new applications for AI gives the impression that the future of entertainment is hard to predict. My general opinion on AI-generated "photographs" has been that they're like early CGI in movies: impressive at a glance but only because we haven't learnt the telltale signs to look for yet. Yet every new version of deep-learning models such as Midjourney produces images with more natural-looking people and more believable surroundings, even if (for now) there's still a general Lynchian vibe, and regularly horrifying mistakes in the fingers and teeth, or objects that float or collide with each other in the wrong ways. The new version of the OpenAI's language model has only been out for a week, and already one user has discovered that it can read and interpret the source code of a video game, and repackage it as a sort of choose-your-own-adventure novel. Who's to say it won't soon be able to create its own games from scratch based on requests? Voice models of the most prominent US celebrities are so easily accessible that creators only need to provide a written script to have audio content of them saying anything.

Building systems that are good for society in the face of complex societal effects requires a dynamic approach. Recent approaches to machine learning (ML) documentation have demonstrated the promise of discursive frameworks for deliberation about these complexities. However, these developments have been grounded in a static ML paradigm, leaving the role of feedback and post-deployment performance unexamined. Meanwhile, recent work in reinforcement learning has shown that the effects of feedback and optimization objectives on system behavior can be wide-ranging and unpredictable. In this paper we sketch a framework for documenting deployed and iteratively updated learning systems, which we call Reward Reports. Taking inspiration from various contributions to the technical literature on reinforcement learning, we outline Reward Reports as living documents that track updates to design choices and assumptions behind what a particular automated system is optimizing for. They are intended to track dynamic phenomena arising from system deployment, rather than merely static properties of models or data. After presenting the elements of a Reward Report, we discuss a concrete example: Meta's BlenderBot 3 chatbot. Several others for game-playing (DeepMind's MuZero), content recommendation (MovieLens), and traffic control (Project Flow) are included in the appendix.

In this study, we present a hybrid CNN-RNN approach to investigate long-term survival of subjects in a lung cancer screening study. Subjects who died of cardiovascular and respiratory causes were identified whereby the CNN model was used to capture imaging features in the CT scans and the RNN model was used to investigate time series and thus global information. The models were trained on subjects who underwent cardiovascular and respiratory deaths and a control cohort matched to participant age, gender, and smoking history. The combined model can achieve an AUC of 0.76 which outperforms humans at cardiovascular mortality prediction. The corresponding F1 and Matthews Correlation Coefficient are 0.63 and 0.42 respectively. The generalisability of the model is further validated on an 'external' cohort. The same models were applied to survival analysis with the Cox Proportional Hazard model. It was demonstrated that incorporating the follow-up history can lead to improvement in survival prediction. The Cox neural network can achieve an IPCW C-index of 0.75 on the internal dataset and 0.69 on an external dataset. Delineating imaging features associated with long-term survival can help focus preventative interventions appropriately, particularly for under-recognised pathologies thereby potentially reducing patient morbidity.

Connected and Automated Vehicles (CAVs) are one of the emerging technologies in the automotive domain that has the potential to alleviate the issues of accidents, traffic congestion, and pollutant emissions, leading to a safe, efficient, and sustainable transportation system. Machine learning-based methods are widely used in CAVs for crucial tasks like perception, motion planning, and motion control, where machine learning models in CAVs are solely trained using the local vehicle data, and the performance is not certain when exposed to new environments or unseen conditions. Federated learning (FL) is an effective solution for CAVs that enables a collaborative model development with multiple vehicles in a distributed learning framework. FL enables CAVs to learn from a wide range of driving environments and improve their overall performance while ensuring the privacy and security of local vehicle data. In this paper, we review the progress accomplished by researchers in applying FL to CAVs. A broader view of the various data modalities and algorithms that have been implemented on CAVs is provided. Specific applications of FL are reviewed in detail, and an analysis of the challenges and future scope of research are presented.

We have entered the brave new world of AI chatbots. This means everything from reenvisioning how students learn in school to protecting ourselves from mass-produced misinformation. It also means heeding the mounting calls to regulate AI to help us navigate an era in which computers write as fluently as people. So far, there is more agreement on the need for AI regulation than on what this would entail. Mira Murati, head of the team that created the chatbot app ChatGPT--the fastest growing consumer-Internet app in history--said governments and regulators should be involved, but she didn't suggest how.

We are the Concept Art Association, an advocacy organization for artists working in entertainment. Our board member, Karla Ortiz, has been one of the leaders in our industry fighting back against the unethical practices happening in the AI text-to-image space. As an organization and as individuals we deeply care about this issue, not just for those actively working as visual artists, but for future generations of artists and for the preservation of our creative industries. A text-to-image model takes input from a user in the form of a natural language prompt and produces an image matching that prompt. To condition that capability the model needs to be trained on a huge collection of images, media, and text descriptions scraped from the web and collected in the form of a "dataset " in order to extract and encode an intricate statistical survey of the dataset's items.

Former U.S. Amb. to NATO Kurt Volker says the Russian fighter jet collision was'intentional' and requires a'firm response' from the U.S. The United States has resumed its normal flights through international waters over the Black Sea following the crash of a drone due to Russian interference. U.S. officials said Friday that a RQ-4 Global Hawk flew through the region -- the first U.S. aircraft to do so since the skirmish, according to Reuters. An RQ-4 Global Hawk takes off from Andersen Air Force Base, Guam (U.S. Air Force photo/Senior Airman Nichelle Anderson) Military officials assured the public that the Russian harassment of the US drone on Tuesday would not affect regular operations in the region. Defense Secretary Lloyd Austin summarized the incident Wednesday in a press conference, saying, "Two Russian jets dumped fuel on an unmanned U.S. MQ-9 aircraft conducting routine operations in international airspace. And one Russian jet intercepted and hit our MQ-9 aircraft, resulting in a crash."

Dr. Robert Marks is a professor at Baylor University. He warns the general public against accepting too much "hype" when it comes to artificial intelligence. According to an expert on artificial intelligence (AI), the biggest threats from the emerging technology include the United States military falling behind other countries, as well as unreliable "woke" bias in Chat GPS. However, Robert J. Marks II, PhD, a professor at Baylor University, hit back against sci-fi warnings of sentient machines and reassured Americans that they won't become "pets" to an all-controlling technology. In an interview with Fox News Digital, Marks, the Director of the Walter Bradley Center for Natural & Artificial Intelligence, suggested that the culture gets a lot wrong about the technology.

The suspension by the BBC of Gary Lineker, a well-known footballer-turned-broadcaster, over a tweet comparing the United Kingdom's new immigration bill with 1930s' Nazi Germany, is exposing the double standards in British journalism and politics. Afghan journalists are paying with their lives in the power struggle between the Taliban and ISIL (ISIS). Producer Flo Phillips looks into the targeted explosion that marked Afghanistan's National Journalism Day. Are you OK with AI? Artificial intelligence is not exactly new, but it is having a blockbuster few months with constantly developing software that can do much more than responding to instructions. Producer Ahmed Madi explains the potential of AI and how it might transform the media you consume.