'A new norm': BBC visits Doha market starting to fill up again two weeks into Iran war At the beginning of the conflict between Israel and the US, and Iran on 28 February, Doha's Souq Waqif market was almost empty, with those in the usually safe and stable capital shocked by the attacks in the region. Qatar's neighbouring countries have felt the impact of Tehran's retaliatory strikes, with at least 18 people killed across the Gulf states so far . Meanwhile, most of the strikes aimed at Qatar - some targeting US military bases - have been intercepted by air defences, with little damage done on the ground and no deaths reported in the country. As the conflict in the Middle East enters its third week, Doha's best-known market is starting to look busy again - and the BBC's Barbara Plett Usher has visited to ask people there how they are feeling. Voiced by Domhnall Gleeson and directed by John Kelly, Retirement Plan is nominated for Best Animated Short Film at the 98th Academy Awards.

Oscars 2026: Here are all the moments you didn't see on TV The 98th Academy Awards featured emotional speeches, comical relief and a bevy of backstage fun. While movie magic plays a role in the show itself (the ceremony, after all, is actually hosted at the Dolby Theatre in a shopping centre), there is a lot you don't see on TV. Frankenstein production designer addressed the media with his Oscar statuette in one hand and what appeared to be a beer in the other and Mr Nobody Against Putin filmmaker Pasha Talankin re-lived his Oscars win by re-reading the envelope that announced that his movie won the award for documentary feature film. We saw some of the tightest security in recent years and witnessed the frenzied panic after one Oscar award became two when those vying for best short action film was announced as a historic tie. Here's what it's like on the scene during Hollywood's biggest night and everything you did not see on TV.

U.S. President Donald Trump speaks to reporters aboard Air Force One on a flight to Washington on Sunday. SAN FRANCISCO - U.S. President Donald Trump on Sunday accused Iran of using artificial intelligence as a "disinformation weapon" to misrepresent its wartime successes and support. "AI can be very dangerous, we have to be very careful with it," Trump said to reporters on Air Force One shortly after he made a post on his Truth Social platform where he accused Western media outlets without evidence of "close coordination" with Iran to spread AI-generated fake news." The comments come amid renewed tensions between the Federal Communications Commission and broadcasters after Trump took aim at media coverage of the U.S. and Israel's war with Iran. FCC Chairman Brendan Carr on Saturday threatened to pull licenses of broadcasters who did not "correct course" on their coverage.

Could Iran be using China's BeiDou system? Footage shows a fire burning near Dubai International Airport after a drone ignited a fuel tank, according to authorities in the UAE. Civil defence crews say the blaze is under control. What is force majeure and why are some Gulf countries invoking it?

Race on to establish globally recognised'AI-free' logo Organisations worldwide are racing to develop a universally recognised label for human-made products and services as part of the growing backlash against AI use. Declarations like Proudly Human, Human-made, 'No A.I and AI-free are appearing across films, marketing, books and websites. It is in response to fears that jobs or entire professions are being swept away in a wave of AI-powered automation. BBC News has counted at least eight different initiatives trying to come up with a label that could get the kind of global recognition that the Fair Trade logo has for ethically made products. But with so many competing labels - as well as confusion over the definition of AI-free - experts say consumers are in danger of being left confused unless a single standard can be agreed on.

'We will go wherever they hide': Rooting out IS in Somalia A figure appears in the picture, moving through a valley. He has been to fetch water for his friends, says the drone operator. He is running and carrying something on his back, adds another soldier. The man on the screen is near a cave, which the army believes is a hideout for 50 to 60 IS fighters. The Puntland Defence Forces have about 500 soldiers stationed at this base in the north-east of Somalia. Ten years ago the barren and inhospitable landscape was home to only a few nomadic communities, but that changed when IS established a foothold here, shifting its focus to Africa as its fighters were driven out of their strongholds in Syria and Iraq.

The first reports appeared on foreign screens, beyond the reach of most Iranians. On 28 February Prime Minister Benjamin Netanyahu said there were signs that the tyrant is no more, suggesting Supreme Leader Ayatollah Ali Khamenei had been killed in a joint US-Israeli strike. Iranians watching state television, however, encountered silence. Government officials would neither confirm nor deny Khamenei's death. On one of the state broadcaster's channels, IRTV3, one news presenter urged viewers to trust him and the latest information the government had.

Offline reinforcement learning (RL) aims to learn decision policies from a fixed batch of logged transitions, without additional environment interaction. Despite remarkable empirical progress, offline RL remains fragile under distribution shifts: value-based methods can overestimate the value of unseen actions, yielding policies that exploit model errors rather than genuine long-term rewards. We propose \emph{Bayesian Conservative Policy Optimization (BCPO)}, a unified framework that converts epistemic uncertainty into \emph{provably conservative} policy improvement. BCPO maintains a hierarchical Bayesian posterior over environment/value models, constructs a \emph{credible lower bound} (LCB) on action values, and performs policy updates under explicit KL regularization toward the behavior distribution. This yields an uncertainty-calibrated analogue of conservative policy iteration in the offline regime. We provide a finite-MDP theory showing that the pessimistic fixed point lower-bounds the true value function with high probability and that KL-controlled updates improve a computable return lower bound. Empirically, we verify the methodology on a real offline replay dataset for the CartPole benchmark obtained via the \texttt{d3rlpy} ecosystem, and report diagnostics that link uncertainty growth and policy drift to offline instability, motivating principled early stopping and calibration

Sparse regularization plays a central role in solving inverse problems arising from incomplete or corrupted measurements. Different regularizers correspond to different prior assumptions about the structure of the unknown signal, and reconstruction performance depends on how well these priors match the intrinsic sparsity of the data. This work investigates the effect of sparsity priors in inverse problems by comparing conventional L1 regularization with the Variational Garrote (VG), a probabilistic method that approximates L0 sparsity through variational binary gating variables. A unified experimental framework is constructed across multiple reconstruction tasks including signal resampling, signal denoising, and sparse-view computed tomography. To enable consistent comparison across models with different parameterizations, regularization strength is swept across wide ranges and reconstruction behavior is analyzed through train-generalization error curves. Experiments reveal characteristic bias-variance tradeoff patterns across tasks and demonstrate that VG frequently achieves lower minimum generalization error and improved stability in strongly underdetermined regimes where accurate support recovery is critical. These results suggest that sparsity priors closer to spike-and-slab structure can provide advantages when the underlying coefficient distribution is strongly sparse. The study highlights the importance of prior-data alignment in sparse inverse problems and provides empirical insights into the behavior of variational L0-type methods across different information bottlenecks.

Reinforcement learning (RL) has become a standard technique for post-training diffusion-based image synthesis models, as it enables learning from reward signals to explicitly improve desirable aspects such as image quality and prompt alignment. In this paper, we propose an online RL variant that reduces the variance in the model updates by sampling paired trajectories and pulling the flow velocity in the direction of the more favorable image. Unlike existing methods that treat each sampling step as a separate policy action, we consider the entire sampling process as a single action. We experiment with both high-quality vision language models and off-the-shelf quality metrics for rewards, and evaluate the outputs using a broad set of metrics. Our method converges faster and yields higher output quality and prompt alignment than previous approaches.