Sudan's Khartoum targeted by RSF drones for third day after airport reopens The paramilitary Rapid Support Forces (RSF) have targeted Sudan's capital Khartoum and its main airport with drones, a day after the first passenger flight in two years landed in the city amid the civil war. The government-aligned Sudanese Armed Forces (SAF) intercepted the drones on Thursday, which caused no damage, a military official who spoke on condition of anonymity told The Associated Press news agency. The RSF and SAF did not immediately acknowledge the attack. The airport has come under repeated drone attacks blamed on the RSF since Tuesday. Al Jazeera's Hiba Morgan said "both sides seem to be stepping up the use of drones, with the RSF using them here in the capital, Khartoum, to target facilities such as the airport".

Blackouts hit Russia's Belgorod as Ukrainian drone attacks surge Residents of Russia's Belgorod region say blackouts, air-raid sirens and the sound of gunfire aimed at incoming Ukrainian drones are becoming increasingly common, as Kyiv retaliates against repeated bombardments of its cities with cross-border strikes of its own. It's so loud and so terrifying, says Nina, a Belgorod resident who asked us to change her name. I was coming back from the clinic when a siren went off. As usual, I received Telegram alerts about a drone attack. Then bursts of automatic gunfire broke out, I ran into a nearby courtyard and tried to hide under an arch, she recalls.

In this work, we establish non-asymptotic Berry-Esseen bounds for linear functionals of online least-squares SGD, thereby providing a Gaussian Central Limit Theorem (CL T) in a growing-dimensional regime. To render the theory practically applicable, we further develop an online variance estimator for the asymptotic variance appearing in the CL T and establish high-probability deviation bounds for this estimator. Stochastic gradient descent [56] is a popular optimization algorithm widely used in data science. It is a stochastic iterative method for minimizing the expected loss function by updating model parameters based on the (stochastic) gradient of the loss with respect to the parameters obtained from a random sample. SGD is widely used for training linear and logistic regression models, support vector machines, deep neural networks, and other such machine learning models on large-scale datasets. Because of its simplicity and effectiveness, SGD has become a staple of modern data science and machine learning, and has been continuously improved and extended to handle more complex scenarios. Despite its wide-spread applicability for prediction and point estimation, quantifying the uncertainty associated with SGD is not well-understood. Indeed, uncertainty quantification is a key component of decision making systems, ensuring the credibility and validity of data-driven findings; see, for e.g., [17], for a concrete medical application where it is not enough to just optimize SGD to obtain prediction performance but is more important to quantify the associated uncertainty.

Understanding the dynamics of feature learning in neural networks (NNs) remains a significant challenge. The work of (Mousavi-Hosseini et al., 2023) analyzes a multiple index teacher-student setting and shows that a two-layer student attains a low-rank structure in its first-layer weights when trained with stochastic gradient descent (SGD) and a strong regularizer. This structural property is known to reduce sample complexity of generalization. Indeed, in a second step, the same authors establish algorithm-specific learning guarantees under additional assumptions. In this paper, we focus exclusively on the structure discovery aspect and study it under weaker assumptions, more specifically: we allow (a) NNs of arbitrary size and depth, (b) with all parameters trainable, (c) under any smooth loss function, (d) tiny regularization, and (e) trained by any method that attains a second-order stationary point (SOSP), e.g.\ perturbed gradient descent (PGD). At the core of our approach is a key $\textit{derandomization}$ lemma, which states that optimizing the function $\mathbb{E}_{\mathbf{x}} \left[g_θ(\mathbf{W}\mathbf{x} + \mathbf{b})\right]$ converges to a point where $\mathbf{W} = \mathbf{0}$, under mild conditions. The fundamental nature of this lemma directly explains structure discovery and has immediate applications in other domains including an end-to-end approximation for MAXCUT, and computing Johnson-Lindenstrauss embeddings.

Neural activity forecasting is central to understanding neural systems and enabling closed-loop control. While deep learning has recently advanced the state-of-the-art in the time series forecasting literature, its application to neural activity forecasting remains limited. To bridge this gap, we systematically evaluated eight probabilistic deep learning models, including two foundation models, that have demonstrated strong performance on general forecasting benchmarks. We compared them against four classical statistical models and two baseline methods on spontaneous neural activity recorded from mouse cortex via widefield imaging. Across prediction horizons, several deep learning models consistently outperformed classical approaches, with the best model producing informative forecasts up to 1.5 seconds into the future. Our findings point toward future control applications and open new avenues for probing the intrinsic temporal structure of neural activity.

Course syllabi set the tone and expectations for courses, shaping the learning experience for both students and instructors. In computing courses, especially those addressing fairness and ethics in artificial intelligence (AI), machine learning (ML), and algorithmic design it is imperative that we understand how approaches to navigating barriers to fair outcomes are being addressed.These expectations should be inclusive, transparent, and grounded in promoting critical thinking. Syllabus analysis offers a way to evaluate the coverage, depth, practices, and expectations within a course. Manual syllabus evaluation, however, is time-consuming and prone to inconsistency. To address this, we developed a justice-oriented scoring rubric and asked a large language model (LLM) to review syllabi through a multi-perspective role simulation. Using this rubric, we evaluated 24 syllabi from four perspectives: instructor, departmental chair, institutional reviewer, and external evaluator. We also prompted the LLM to identify thematic trends across the courses. Findings show that multi-perspective evaluation aids us in noting nuanced, role-specific priorities, leveraging them to fill hidden gaps in curricula design of AI/ML and related computing courses focused on fairness and ethics. These insights offer concrete directions for improving the design and delivery of fairness, ethics, and justice content in such courses.

Online shopping giant Temu has agreed to work with the greeting card industry to remove copied designs from its site more quickly. Designers told the BBC the process for getting the plagiarised listings removed has been like the fairground game'whack-a-mole' with copied products re-appearing within days. Temu said protecting intellectual property was a top priority and that it was encouraging sellers to join the trial of a new takedown process specifically for the greetings card industry. Amanda Mountain, the co-founder of York-based Lola Design, discovered the catalogue of designs she had built up over a decade had nearly all been copied. She found the images she had created had been lifted and were being advertised by other sellers on cards and other products like t-shirts.

US to announce'substantial' Russia sanctions The US government will impose a substantial pickup in sanctions against Russia as the fighting in Ukraine continues, according to US Treasury Secretary Scott Bessent. Bessent's comments came just before Nato Secretary-General Mark Rutte was due at the White House, in which he said he hopes to discuss how to deliver Trump's vision of peace in the conflict. Earlier in the day, Rutte said he believes that Trump is the only one who can get this done. At least seven people were killed, including two children, during intense Russian drone and missile strikes on Ukraine - just hours after Trump said plans for a meeting with Vladimir Putin in Budapest had been shelved. Bessent provided no further details on the incoming sanctions, but said they would be announced either after the close this afternoon or first thing tomorrow morning.

Conservative activist Robby Starbuck sued Google, alleging that the tech giant's artificial intelligence systems generated "outrageously false" information about him. On Wednesday, Starbuck said in the lawsuit, filed in Delaware state court, that Google's AI systems falsely called him a "child rapist," "serial sexual abuser" and "shooter" in response to user queries and delivered defamatory statements to millions of users. "Hallucinations are a well-known issue for all LLMs, which we disclose and work hard to minimise," Castaneda said. "But as everyone knows, if you're creative enough, you can prompt a chatbot to say something misleading." Starbuck is best known for opposing diversity, equity and inclusion initiatives.

How successful is Ukraine's'gas war' against Russia? How will Putin travel to Hungary with an ICC arrest warrant? How much of Europe's oil still comes from Russia? A bystander filmed the moment a Russian drone struck near an apartment building in Sumy, Ukraine, injuring several people. Russia launched a wave of deadly drone and missile attacks on cities across Ukraine on Tuesday.