Open-source multimodal large language models (MLLMs) have shown significant potential in a broad range of multimodal tasks. However, their reasoning capabilities remain constrained by existing instruction-tuning datasets, which were predominately repurposed from academic datasets such as VQA, AI2D, and ChartQA. These datasets target simplistic tasks, and only provide phrase-level answers without any intermediate rationales. To address these challenges, we introduce a scalable and cost-effective method to construct a large-scale multimodal instruction-tuning dataset with rich intermediate rationales designed to elicit CoT reasoning. Using only open models, we create a dataset containing 12M instruction-response pairs to cover diverse, reasoning-intensive tasks with detailed and faithful rationales. Experiments demonstrate that training MLLMs on this dataset significantly improves reasoning capabilities, achieving state-of-the-art performance on benchmarks such as MathVerse (+8.1%), MMMU-Pro (+7%), and MuirBench (+13.3%). Additionally, the model demonstrates notable improvements of up to 4% on non-reasoning-based benchmarks. Ablation studies further highlight the importance of key components, such as rewriting and self-filtering, in the dataset construction process.

Many existing jailbreak techniques rely on solving discrete combinatorial optimization, while more recent approaches involve training LLMs to generate multiple adversarial prompts. However, both approaches require significant computational resources to produce even a single adversarial prompt. We hypothesize that the inefficiency of current approaches stems from an inadequate characterization of the jailbreak problem. To address this gap, we formulate the jailbreak problem in terms of alignment. By starting from an available safety-aligned model, we leverage an unsafe reward to guide the safe model towards generating unsafe outputs using alignment techniques (e.g., reinforcement learning from human feedback), effectively performing jailbreaking via alignment. We propose a novel jailbreak method called LIAR (LeveragIng Alignment to jailbReak). To demonstrate the simplicity and effectiveness of our approach, we employ a best-of-N method to solve the alignment problem. LIAR offers significant advantages: lower computational requirements without additional training, fully black-box operation, competitive attack success rates, and more human-readable prompts. We provide theoretical insights into the possibility of jailbreaking a safety-aligned model, revealing inherent vulnerabilities in current alignment strategies for LLMs. We also provide sub-optimality guarantees for the proposed \algo. Experimentally, we achieve ASR comparable to the SoTA with a 10x improvement to perplexity and a Time-to-Attack measured in seconds rather than tens of hours.

Infrared (IR) imaging is commonly used in various scenarios, including autonomous driving, fire safety and defense applications. Thus, semantic segmentation of such images is of great interest. However, this task faces several challenges, including data scarcity, differing contrast and input channel number compared to natural images, and emergence of classes not represented in databases in certain scenarios, such as defense applications. Few-shot segmentation (FSS) provides a framework to overcome these issues by segmenting query images using a few labeled support samples. However, existing FSS models for IR images require paired visible RGB images, which is a major limitation since acquiring such paired data is difficult or impossible in some applications. In this work, we develop new strategies for FSS of IR images by using generative modeling and fusion techniques. To this end, we propose to synthesize auxiliary data to provide additional channel information to complement the limited contrast in the IR images, as well as IR data synthesis for data augmentation. Here, the former helps the FSS model to better capture the relationship between the support and query sets, while the latter addresses the issue of data scarcity. Finally, to further improve the former aspect, we propose a novel fusion ensemble module for integrating the two different modalities. Our methods are evaluated on different IR datasets, and improve upon the state-of-the-art (SOTA) FSS models.

Much effort has been made to explain and improve the success of transfer-based attacks (TBA) on black-box computer vision models. This work provides the first attempt at a priori prediction of attack success by identifying the presence of vulnerable features within target models. Recent work by Chen and Liu (2024) proposed the manifold attack model, a unifying framework proposing that successful TBA exist in a common manifold space. Our work experimentally tests the common manifold space hypothesis by a new methodology: first, projecting feature vectors from surrogate and target feature extractors trained on ImageNet onto the same low-dimensional manifold; second, quantifying any observed structure similarities on the manifold; and finally, by relating these observed similarities to the success of the TBA. We find that shared feature representation moderately correlates with increased success of TBA (\r{ho}= 0.56). This method may be used to predict whether an attack will transfer without information of the model weights, training, architecture or details of the attack. The results confirm the presence of shared feature representations between two feature extractors of different sizes and complexities, and demonstrate the utility of datasets from different target domains as test signals for interpreting black-box feature representations.

Global Climate Models (GCMs) are numerical models that simulate complex physical processes within the Earth's climate system and are essential for understanding and predicting climate change. However, GCMs suffer from systemic biases due to simplifications made to the underlying physical processes. GCM output therefore needs to be bias corrected before it can be used for future climate projections. Most common bias correction methods, however, cannot preserve spatial, temporal, or inter-variable dependencies. We propose a new semi-parametric conditional density estimation (SPCDE) for density correction of the joint distribution of daily precipitation and maximum temperature data obtained from gridded GCM spatial fields. The Vecchia approximation is employed to preserve dependencies in the observed field during the density correction process, which is carried out using semi-parametric quantile regression. The ability to calibrate joint distributions of GCM projections has potential advantages not only in estimating extremes, but also in better estimating compound hazards, like heat waves and drought, under potential climate change. Illustration on historical data from 1951-2014 over two 5x5 spatial grids in the US indicate that SPCDE can preserve key marginal and joint distribution properties of precipitation and maximum temperature, and predictions obtained using SPCDE are better calibrated compared to predictions using asynchronous quantile mapping and canonical correlation analysis, two commonly used bias correction approaches.

The tragedy of the commons illustrates a fundamental social dilemma where individual rational actions lead to collectively undesired outcomes, threatening the sustainability of shared resources. Strategies to escape this dilemma, however, are in short supply. In this study, we explore how artificial intelligence (AI) agents can be leveraged to enhance cooperation in public goods games, moving beyond traditional regulatory approaches to using AI as facilitators of cooperation. We investigate three scenarios: (1) Mandatory Cooperation Policy for AI Agents, where AI agents are institutionally mandated always to cooperate; (2) Player-Controlled Agent Cooperation Policy, where players evolve control over AI agents' likelihood to cooperate; and (3) Agents Mimic Players, where AI agents copy the behavior of players. Using a computational evolutionary model with a population of agents playing public goods games, we find that only when AI agents mimic player behavior does the critical synergy threshold for cooperation decrease, effectively resolving the dilemma. This suggests that we can leverage AI to promote collective well-being in societal dilemmas by designing AI agents to mimic human players.

A significant challenge for autonomous cyber defence is ensuring a defensive agent's ability to generalise across diverse network topologies and configurations. This capability is necessary for agents to remain effective when deployed in dynamically changing environments, such as an enterprise network where devices may frequently join and leave. Standard approaches to deep reinforcement learning, where policies are parameterised using a fixed-input multi-layer perceptron (MLP) expect fixed-size observation and action spaces. In autonomous cyber defence, this makes it hard to develop agents that generalise to environments with network topologies different from those trained on, as the number of nodes affects the natural size of the observation and action spaces. To overcome this limitation, we reframe the problem of autonomous network defence using entity-based reinforcement learning, where the observation and action space of an agent are decomposed into a collection of discrete entities. This framework enables the use of policy parameterisations specialised in compositional generalisation. We train a Transformer-based policy on the Yawning Titan cyber-security simulation environment and test its generalisation capabilities across various network topologies. We demonstrate that this approach significantly outperforms an MLP-based policy when training across fixed-size networks of varying topologies, and matches performance when training on a single network. We also demonstrate the potential for zero-shot generalisation to networks of a different size to those seen in training. These findings highlight the potential for entity-based reinforcement learning to advance the field of autonomous cyber defence by providing more generalisable policies capable of handling variations in real-world network environments.

Scaling robot learning requires data collection pipelines that scale favorably with human effort. In this work, we propose Crowdsourcing and Amortizing Human Effort for Real-to-Sim-to-Real(CASHER), a pipeline for scaling up data collection and learning in simulation where the performance scales superlinearly with human effort. The key idea is to crowdsource digital twins of real-world scenes using 3D reconstruction and collect large-scale data in simulation, rather than the real-world. Data collection in simulation is initially driven by RL, bootstrapped with human demonstrations. As the training of a generalist policy progresses across environments, its generalization capabilities can be used to replace human effort with model generated demonstrations. This results in a pipeline where behavioral data is collected in simulation with continually reducing human effort. We show that CASHER demonstrates zero-shot and few-shot scaling laws on three real-world tasks across diverse scenarios. We show that CASHER enables fine-tuning of pre-trained policies to a target scenario using a video scan without any additional human effort. See our project website: https://casher-robot-learning.github.io/CASHER/

This study presents the conditional neural fields for reduced-order modeling (CNF-ROM) framework to approximate solutions of parametrized partial differential equations (PDEs). The approach combines a parametric neural ODE (PNODE) for modeling latent dynamics over time with a decoder that reconstructs PDE solutions from the corresponding latent states. We introduce a physics-informed learning objective for CNF-ROM, which includes two key components. First, the framework uses coordinate-based neural networks to calculate and minimize PDE residuals by computing spatial derivatives via automatic differentiation and applying the chain rule for time derivatives. Second, exact initial and boundary conditions (IC/BC) are imposed using approximate distance functions (ADFs) [Sukumar and Srivastava, CMAME, 2022]. However, ADFs introduce a trade-off as their second- or higher-order derivatives become unstable at the joining points of boundaries. To address this, we introduce an auxiliary network inspired by [Gladstone et al., NeurIPS ML4PS workshop, 2022]. Our method is validated through parameter extrapolation and interpolation, temporal extrapolation, and comparisons with analytical solutions.

Alignment is a social phenomenon wherein individuals share a common goal or perspective. Mirroring, or mimicking the behaviors and opinions of another individual, is one mechanism by which individuals can become aligned. Large scale investigations of the effect of mirroring on alignment have been limited due to the scalability of traditional experimental designs in sociology. In this paper, we introduce a simple computational framework that enables studying the effect of mirroring behavior on alignment in multi-agent systems. We simulate systems of interacting large language models in this framework and characterize overall system behavior and alignment with quantitative measures of agent dynamics. We find that system behavior is strongly influenced by the range of communication of each agent and that these effects are exacerbated by increased rates of mirroring. We discuss the observed simulated system behavior in the context of known human social dynamics.