Standard Bayesian Optimization (BO) assumes uniform smoothness across the search space an assumption violated in multi-regime problems such as molecular conformation search through distinct energy basins or drug discovery across heterogeneous molecular scaffolds. A single GP either oversmooths sharp transitions or hallucinates noise in smooth regions, yielding miscalibrated uncertainty. We propose RAMBO, a Dirichlet Process Mixture of Gaussian Processes that automatically discovers latent regimes during optimization, each modeled by an independent GP with locally-optimized hyperparameters. We derive collapsed Gibbs sampling that analytically marginalizes latent functions for efficient inference, and introduce adaptive concentration parameter scheduling for coarse-to-fine regime discovery. Our acquisition functions decompose uncertainty into intra-regime and inter-regime components. Experiments on synthetic benchmarks and real-world applications, including molecular conformer optimization, virtual screening for drug discovery, and fusion reactor design, demonstrate consistent improvements over state-of-the-art baselines on multi-regime objectives.

The model is trained to minimise the value function while still accurately predicting the transitions in the dataset, forcing the policy to act conservatively in areas not covered by the dataset. To approximately solve the two-player game, we alternate between optimising the policy and adversarially optimising the model.

Large Language Models (LLMs) have revolutionized numerous applications, making them an integral part of our digital ecosystem. However, their reliability becomes critical, especially when these models are exposed to misinformation. We primarily analyze the susceptibility of state-of-the-art LLMs to factual inaccuracies when they encounter false information in a QnA scenario, an issue that can lead to a phenomenon we refer to as *knowledge drift*, which significantly undermines the trustworthiness of these models. We evaluate the factuality and the uncertainty of the models' responses relying on Entropy, Perplexity, and Token Probability metrics. Our experiments reveal that an LLM's uncertainty can increase up to 56.6% when the question is answered incorrectly due to the exposure to false information. At the same time, repeated exposure to the same false information can decrease the models uncertainty again (-52.8% w.r.t. the answers on the untainted prompts), potentially manipulating the underlying model's beliefs and introducing a drift from its original knowledge. These findings provide insights into LLMs' robustness and vulnerability to adversarial inputs, paving the way for developing more reliable LLM applications across various domains. The code is available at https://github.com/afastowski/knowledge_drift.

Offline reinforcement learning (RL) aims to find performant policies from logged data without further environment interaction. Model-based algorithms, which learn a model of the environment from the dataset and perform conservative policy optimisation within that model, have emerged as a promising approach to this problem. In this work, we present Robust Adversarial Model-Based Offline RL (RAMBO), a novel approach to model-based offline RL. We formulate the problem as a two-player zero sum game against an adversarial environment model. The model is trained to minimise the value function while still accurately predicting the transitions in the dataset, forcing the policy to act conservatively in areas not covered by the dataset. To approximately solve the two-player game, we alternate between optimising the policy and adversarially optimising the model. The problem formulation that we address is theoretically grounded, resulting in a probably approximately correct (PAC) performance guarantee and a pessimistic value function which lower bounds the value function in the true environment. We evaluate our approach on widely studied offline RL benchmarks, and demonstrate that it outperforms existing state-of-the-art baselines.

An expert on the digital afterlife responds to Cat Rambo's "The Woman Who Wanted to Be Trees." You never know precisely how much time you have left, despite what life insurance industry mortality tables or death-prediction startups might claim. Now, an emerging field of death tech is capitalizing on such anxiety by pitching individual immortality as deepfakes or AI-driven chatbots. Meanwhile, we're facing an ongoing environmental catastrophe perpetrated by colonialism and relentless extraction. These two forms of existential uncertainty may seem separate--but they are intrinsically related.

TORONTO, ONTARIO, CANADA - 2015/03/29: The Bank of Montreal, or BMO Financial Group, is one of the ... [ ] Big Five banks in Canada. BMO Financial Group, the over 200 year old banking group also known as Bank of Montreal, is the 8th largest bank, by assets, in North America and one of the'Big Five' Canadian banks. After exploring intelligent automation for several years now, BMO is accelerating its robotic process automation (RPA) strategy. It is building on the strong foundation of RPA, machine learning, and AI to enable a future of digital and human workforce collaboration. These capabilities been evolving rapidly since 2017, when Randy Bean and I first wrote about BMO's work in the automation space.

New details about Apple's iPhone 8 surfaced over the weekend, and they appear to suggest that the device could come with a new feature designed to activate Siri in the absence of the physical Home button. As per earlier reports, iPhone 8 is going to launch sans a physical Home button due to its all-screen design. Now because the device is unlikely to sport a physical key on its front, consumers are left with no other option than to activate Apple's voice assistant using the "Hey Siri" voice command. For those who do not like resorting to the voice command to wake Siri, Apple could be preparing a new alternative. This feature will reportedly enable users to activate Siri by holding down the Sleep/Wake button of the handset.

If you are a fan of Siri, then good news: Apple's voice assistant looks like it will be getting a promotion in the near future, thanks to the combination of iOS 11 and the "D22" iPhone (the leaked model thought to be the forthcoming "iPhone Pro, 8 or X"). If you're not a fan of Siri, well, look away now – some of this isn't pretty. For the first time, Siri will sync across devices, letting details it has picked up from your phone affect how it answers on your iPad or computer. The service is also seeing a slight refocus: Siri isn't just a voice assistant anymore, since "Siri" will also be responsible for suggesting topics in Apple News based on your web browsing, or for suggesting a calendar event if you make a booking online. It'll also be able to do translations for the first time, from English (only) to five languages including French and Chinese.

More details have been revealed about the highly anticipated iPhone 8 by software leaks from Apple, including the device's ability to mute notifications when it detects you are looking at it. While the name and precise release date of the next big change to the Apple's iPhone, dubbed the D22 iPhone, is unknown the leaked software for Apple's upcoming HomePod smart speaker has already revealed what the phone will look like, that it will have face recognition and other details. Now further analysis by Brazilian Apple site iHelp BR has found suggestions that the iPhone 8's face recognition feature is designed to work when the smartphone is flat on a desk and doesn't require the device to be held up at head height to unlock the phone with a face. One of the criticisms of current face and iris recognition technology, as employed by the Samsung Galaxy S8 and Microsoft's Windows Hello-enabled Surface computers, is that it must be either brought up to eye level to work or must be angled in such a way that it can clearly see the whole of your face. Should Apple have designed a way to allow the infra-red based face recognition system to work at a much wider angle, including flat on a desk, it could go some way to alleviating the issues plaguing current iterations of the technology.