Fine-tuning text-to-image diffusion models with human feedback is an effective method for aligning model behavior with human intentions. However, this alignment process often suffers from slow convergence due to the large size and noise present in human feedback datasets. In this work, we propose FiFA, a novel automated data filtering algorithm designed to enhance the fine-tuning of diffusion models using human feedback datasets with direct preference optimization (DPO). Specifically, our approach selects data by solving an optimization problem to maximize three components: preference margin, text quality, and text diversity. The concept of preference margin is used to identify samples that contain high informational value to address the noisy nature of feedback dataset, which is calculated using a proxy reward model. Additionally, we incorporate text quality, assessed by large language models to prevent harmful contents, and consider text diversity through a k-nearest neighbor entropy estimator to improve generalization. Finally, we integrate all these components into an optimization process, with approximating the solution by assigning importance score to each data pair and selecting the most important ones. As a result, our method efficiently filters data automatically, without the need for manual intervention, and can be applied to any large-scale dataset. Experimental results show that FiFA significantly enhances training stability and achieves better performance, being preferred by humans 17% more, while using less than 0.5% of the full data and thus 1% of the GPU hours compared to utilizing full human feedback datasets. Warning: This paper contains offensive contents that may be upsetting. Large-scale models trained on extensive web-scale datasets using diffusion techniques (Ho et al., 2020; Song et al., 2020), such as Stable Diffusion (Rombach et al., 2022), Dall-E (Ramesh et al., 2022), and Imagen (Saharia et al., 2022), have enabled the generation of high-fidelity images from diverse text prompts. However, several failure cases remain, such as difficulties in illustrating text content, incorrect counting, or insufficient aesthetics for certain text prompts (Lee et al., 2023; Fan et al., 2024; Black et al., 2023). Fine-tuning text-to-image diffusion models using human feedback has recently emerged as a powerful approach to address this issue (Black et al., 2023; Fan et al., 2024; Prabhudesai et al., 2023; Clark et al., 2023). Unlike the conventional optimization strategy of likelihood maximization, this framework first trains reward models using human feedback (Kirstain et al., 2024; Wu et al., 2023; Xu et al., 2024) and then fine-tunes the diffusion models to maximize reward scores through policy gradient (Fan et al., 2024; Black et al., 2023) or reward-gradient based techniques (Prabhudesai et al., 2023; Clark et al., 2023).

The Canadian women's soccer team was dealt a heavy blow Saturday after FIFA announced the women's national team would be deducted six points from the standings in the Paris Olympics after staffers were caught using drones to spy on New Zealand during closed-door training sessions. Following its investigation, the FIFA Appeal Committee announced the Canadian Soccer Association was responsible for failing to ensure its staff members were in compliance with Olympic rules. "CSA was found responsible for failing to respect the applicable FIFA regulations in connection with its failure to ensure the compliance of its participating officials of the Games of the XXXIII Olympiad Paris 2024 Final Competition (OFT) with the prohibition on flying drones over any training sites," the statement said. "The officials were each found responsible for offensive behavior and violation of the principles of fair play in connection with the CSA's Women's representative team's drones usage in the scope of the OFT." Head coach Bev Priestman was removed from her position Thursday night after two staff members were sent home from Paris when an investigation found that analyst Joseph Lombardi had allegedly used a drone to spy on New Zealand's practice sessions.

FIFA deducted six points from Canada in the Paris Olympics women's football tournament and banned three coaches for one year each in a drone spying scandal. The stunning swath of punishments, announced late on Saturday, includes a 200,000-Swiss-franc ( 226,000) fine for the Canadian football federation in a case that has spiralled at the Summer Games. Two assistant coaches were caught using drones to spy on opponent New Zealand's practices before their opening game on Wednesday. Head coach Bev Priestman, who led Canada to the Olympic title in Tokyo in 2021, already was suspended by the national football federation and then removed from the Olympic tournament. She is now banned from all football by FIFA for one year.

Football, like everything else important in life, is about stories. People implant themselves into the narrative: where they were when they saw Maradona's handball, the strangers they hugged when Ole Gunnar Solskjær scored that historic last-minute winner at the 1999 Champions League final. No doubt new tales are already being conjured around Jude Bellingham's scissor kick against Slovakia in the dying seconds of Sunday's Euro 24 match. Sport is a nostalgia machine – and this is as true for video game simulations as it is for the real thing. Every gamer has their favourite footie sim, but for me, and many other players of my … ahem, vintage … it was Pro Evolution Soccer, numbers 3 to 6. This was the early 2000s, the age of the PlayStation 2. I was a writer for hire at Future Publishing, basically hanging out at its office in Bath, working mostly on the Official PlayStation magazine.

Two years ago, the long and lucrative relationship between Electronic Arts and Fifa broke down, with EA taking its ball home and launching EA Sports FC, a new brand for its footie sim series. Fifa president Gianni Infantino made a sulky declaration that he would find a new developer and that, "the only authentic, real game that has the Fifa name will be the best one available for gamers and football fans". This seemed like a ludicrous boast: EA had 20 years of experience making mainstream football sims – an expensive and highly sophisticated endeavour. How could Fifa hope to find a studio capable of competing? Well, it looks as if the global football body may have found its new best friend.

Deep learning methods have led to significant improvements in the performance on the facial landmark detection (FLD) task. However, detecting landmarks in challenging settings, such as head pose changes, exaggerated expressions, or uneven illumination, continue to remain a challenge due to high variability and insufficient samples. This inadequacy can be attributed to the model's inability to effectively acquire appropriate facial structure information from the input images. To address this, we propose a novel image augmentation technique specifically designed for the FLD task to enhance the model's understanding of facial structures. To effectively utilize the newly proposed augmentation technique, we employ a Siamese architecture-based training mechanism with a Deep Canonical Correlation Analysis (DCCA)-based loss to achieve collective learning of high-level feature representations from two different views of the input images. Furthermore, we employ a Transformer + CNN-based network with a custom hourglass module as the robust backbone for the Siamese framework. Extensive experiments show that our approach outperforms multiple state-of-the-art approaches across various benchmark datasets.

Welcome back to Pushing Buttons! First up – last week's newsletter had a few errors in it. Most obviously, I referred to the Meta Quest 2 headset as the now-discontinued Oculus Go (even though I'd just been playing with the Quest 2, to compare it with PSVR2 – nice job, brain). I also gave some incorrect pricing info. A corrected version is on the Guardian site.

FIFA's official partner Adidas came up with an undoubtedly original idea for this year's ball. Their priorities were definitely to implement this year's buzzwords – technology, IoT, AI, and machine learning, but at the same time to include sustainability and upcycling. Now, were these trends used for a purpose and how were they justified? The ball is designed as an homage to Qatar's architecture, national flag, and traditional boats, while the name itself translates as'Journey'. The main feature of this ball is its lightness and speed. More precisely, "the highest level of accuracy" when compared to previous balls.

I would love to tell you that I was first introduced to dance music in underground Berlin clubs, where mysterious resident DJs blew my teenage mind performing indescribable magic with beats and synth lines. But that would be a lie. My first introduction to dance music came in the form of a futuristic 90s racing game called WipEout. Playing obsessively at a friend's house, I was introduced to the Chemical Brothers and Orbital, who both graced the soundtrack; not long after, the admirably chaotic sim Crazy Taxi introduced me to the Offspring, and Tony Hawk's Pro Skater had me grinding around to Bad Religion. I first heard Garbage on the soundtrack of an obscure PlayStation 2 DJ game, 2003's Amplitude, made by a Boston developer called Harmonix – the same developer that would later go on to create the insanely popular Guitar Hero series.

There are many calls made by referees that are still debated by fans even to this day. This includes the controversial goal given to George Hurst in the 1966 World Cup Final, which allowed him to score a hat trick. Who can forget Diego Maradona's famous handball in 1986, which resulted in a goal against England? FIFA is trying to not only reduce such infamous moments but to help out the referees who often do not have a clear sight of what's going on. This is why the officials at FIFA have been experimenting with new AI technology that can track player motions and allow the referees to make more accurate offside calls.