Soon-to-be-laid-off Meta contractors say they're being treated differently than Mark Zuckerberg's full-time employees, who stand to receive more generous severance packages. Now we're being left behind," chanted a horde of contract workers who gathered outside Meta's offices in Dublin, Ireland, on Friday afternoon. Waving flags, brandishing signs, and armed with whistles and vuvuzelas, they were out to protest a round of planned layoffs. The workers are employed by Dublin-based company Covalen, which handles content moderation and data labeling services that help Meta to fine-tune its AI products. In April, Covalen told 700 employees that their jobs were at risk, citing "reduced demand," WIRED reported . A large swath of the affected workers won't receive any severance because they've been employed for less than two years. The rest are being offered the minimum payout required under local labor laws--two weeks' pay for every year of employment--according to the Communications Workers' Union (CWU), whose members include Covalen employees. "We're just getting the crumbs here," Aadel Obaid, a team manager at Covalen who is part of the planned layoffs, tells WIRED. "Give us a little bit of the pie." To try to compel Covalen into revising the severance package, workers voted to strike outside the company's corporate office, before marching to Meta's nearby European headquarters. According to John Bohan, an organizer at the CWU, Meta could use its leverage as an anchor client to pressure Covalen into offering its employees an enhanced severance package. The workers are asking for double what's currently being offered--and at least some form of payment for workers who don't meet the two-year threshold. The company could also release Covalen workers from a "cooldown period" preventing them from working on another Meta account for six months after being laid off, Bohan says. At 1 pm local time on Friday, the striking workers began to gather outside Covalen's corporate headquarters, a red-brick office building on an otherwise largely residential street in the heart of Dublin. The protests began with a wall of sound: the workers beat drums, booed, whistled, shouted, and catcalled. Then came a volley of call-and-response chants led by a worker with a megaphone. The building's security guard watched, bemused, from inside the lobby, hands on his hips. Two hours later, the group--now more than 150 people--began to march down the center of the mile-long stretch of road to Meta's campus, slowing the trailing traffic to a crawl. Dubliners enjoying the early onset of summer stopped to gawp; some applauded. When the protesters arrived at Meta's complex, two security guards stood with crossed arms, blocking the way. The group set up at the gates and began another round of chants: "We scrub the feed.

We study Online Lazy Gradient Descent for optimisation on a strongly convex domain. The algorithm is known to achieve O( N) regret against adversarial opponents; here we show it is universal in the sense that it also achieves O(log N) expected regret against i.i.d opponents. This improves upon the more complex metaalgorithm of Huang et al [20] that only gets O( Nlog N) and O(log N) bounds. In addition we show that, unlike for the simplex, order bounds for pseudo-regret and expected regret are equivalent for strongly convex domains.

In optimization problems, some variable subsets may have a joint non-linear or non-monotonical influence on the function value. Therefore, knowledge of variable dependencies may be crucial for effective optimization, and many state-of-the-art optimizers leverage it to improve performance. However, some real-world problem instances may be the subject of noise of various origins. In such a case, variable dependencies relevant to optimization may be hard or impossible to tell using dependency checks sufficient for problems without noise, making highly effective operators, e.g., Partition Crossover (PX), useless. Therefore, we use Statistical Linkage Learning (SLL) to decompose problems with noise and propose a new SLL-dedicated mask construction algorithm. We prove that if the quality of the SLL-based decomposition is sufficiently high, the proposed clustering algorithm yields masks equivalent to PX masks for the noise-free instances. The experiments show that the optimizer using the proposed mechanisms remains equally effective despite the noise level and outperforms state-of-the-art optimizers for the problems with high noise.

While test-time scaling has enabled large language models to solve highly difficult tasks, state-of-the-art results come at exorbitant compute costs. These inefficiencies can be attributed to the miscalibration of post-trained language models, and the lack of calibration in popular sampling techniques. Here, we present Online Reasoning Calibration (ORCA), a framework for calibrating the sampling process that draws upon conformal prediction and test-time training. Specifically, we introduce a meta-learning procedure that updates the calibration module for each input. This allows us to provide valid confidence estimates under distributional shift, e.g. in thought patterns that occur across different stages of reasoning, or in prompt distributions between model development and deployment. ORCA not only provides theoretical guarantees on conformal risks, but also empirically shows higher efficiency and generalization across different reasoning tasks. At risk level $δ=0.1$, ORCA improves Qwen2.5-32B efficiency on in-distribution tasks with savings up to 47.5% with supervised labels and 40.7% with self-consistency labels. Under zero-shot out-of-domain settings, it improves MATH-500 savings from 24.8% of the static calibration baseline to 67.0% while maintaining a low empirical error rate, and the same trend holds across model families and downstream benchmarks. Our code is publicly available at https://github.com/wzekai99/ORCA.

Striking an optimal balance between predictive performance and fairness continues to be a fundamental challenge in machine learning. In this work, we propose a post-processing framework that facilitates fairness-aware prediction by leveraging model ensembling. Designed to operate independently of any specific model internals, our approach is widely applicable across various learning tasks, model architectures, and fairness definitions. Through extensive experiments spanning classification, regression, and survival analysis, we demonstrate that the framework effectively enhances fairness while maintaining, or only minimally affecting, predictive accuracy.

Vision and diverse languages are important information sources in our living world.

In this paper, we establish an upper bound on the return gap between the oracle expert policy and an optimal decision tree policy. This enables us to recast the DT extraction problem into a novel non-euclidean clustering problem over the local observation and action values space of each agent, with action values as cluster labels and the upper bound on the return gap as clustering loss.