While scaling laws govern aggregate large language model performance, no scaling law has linked factual recall to both model size and training-data composition. We evaluated 38 models on over 8,900 scholarly references evaluated by an automated reference verification system. Recall quality follows a sigmoid in the log-linear combination of model parameter count and topic representation in training data. These two variables alone explain 60% of the variance across 16 dense models from four families, rising to 74-94% within individual families. The form matches a superposition-inspired account in which recall is gated by a signal-to-noise ratio: signal strength scales with concept frequency and the noise floor with model capacity.

One of the core challenges frequently cited in the multi-agent reinforcement learning (MARL) literature motivating the framing of a sequential decision-making problem as a multi-agent problem, instead of a centralised single-agent problem, is the exponential growth in the action space with the number of agents. The assumption that this is always a challenge suggests that this exponentially larger action space poses two specific problems compared with centralised approaches: (1) overwhelming memory requirements and (2) low sample efficiency due to the large optimisation space. Although a core tenet within the MARL community, few works have concretely tested this assumption empirically within a controlled setting to give some indication of its severity in practice. In this work, we compare fully centralised learning with fully decentralised learning. Using a novel N-agent array game akin to the canonical Climbing matrix game, we re-establish a well-known result; that fully centralised learning is able to find the globally optimal solution while decentralised learning fails. We further demonstrate that these trends hold for more modern MARL benchmarks that run on hardware accelerators and leverage the computational efficiency gains of the JAX framework.

We investigate the impact of internal spin on chaos in billiard systems. Extending the standard point-particle billiard by coupling translational and rotational degrees of freedom through a dimensionless spin parameter α= I/(mr^2) ∈ [0,1], we find that spin reduces chaos monotonically but does not eliminate it. In the Bunimovich stadium and Sinai billiard, the Lyapunov exponent decreases with α but remains positive throughout the physical range, while the circle and rectangle remain integrable. Finite-time Lyapunov exponent distributions reveal a mixed phase space in which spin creates islands of regularity while the majority of trajectories remain chaotic. The mechanism is a conserved quantity Q = v_∥ - αu preserved through each collision, which constrains the dynamics on sequences of same-orientation wall collisions and explains why spin suppresses chaos more effectively in geometries with longer flat sections. We further show that the Datseris–Hupe–Fleischmann scaling λ∝ 1/f_ chaotic fails for spinning billiards: spin reduces the intensity of chaos, not merely the fraction of chaotic trajectories.

Active Inference is a recently developed framework for modeling decision processes under uncertainty. Over the last several years, empirical and theoretical work has begun to evaluate the strengths and weaknesses of this approach and how it might be extended and improved. One recent extension is the "sophisticated inference" (SI) algorithm, which improves performance on multi-step planning problems through a recursive decision tree search. However, little work to date has been done to compare SI to other established planning algorithms in reinforcement learning (RL). In addition, SI was developed with a focus on inference as opposed to learning. The present paper therefore has two aims. First, we compare performance of SI to Bayesian RL schemes designed to solve similar problems. Second, we present and compare an extension of SI - sophisticated learning (SL) - that more fully incorporates active learning during planning. SL maintains beliefs about how model parameters would change under the future observations expected under each policy. This allows a form of counterfactual retrospective inference in which the agent considers what could be learned from current or past observations given different future observations. To accomplish these aims, we make use of a novel, biologically inspired environment that requires an optimal balance between goal-seeking and active learning, and which was designed to highlight the problem structure for which SL offers a unique solution. This setup requires an agent to continually search an open environment for available (but changing) resources in the presence of competing affordances for information gain. Our simulations demonstrate that SL outperforms all other algorithms in this context - most notably, Bayes-adaptive RL and upper confidence bound (UCB) algorithms, which aim to solve multi-step planning problems using similar principles (i.e., directed exploration and counterfactual reasoning about belief updates given different possible actions/observations). These results provide added support for the utility of Active Inference in solving this class of biologically-relevant problems and offer added tools for testing hypotheses about human cognition.

We investigate a strongly coupled finite-density anisotropic fluid in 2 + 1 dimensions dual to an asymptotically AdS black brane that is a solution of Einstein-Maxwell-Axion theory in 3 + 1 dimensions. Despite the anisotropy, the fluid thermodynamic properties align with those of a conformal fluid. Moreover, we show that the fluid is stable under the increase of the anisotropy parameter. Additionally, we analyse the DC conductivity of the anisotropic fluid, showing its compatibility with momentum dissipation due to translational symmetry breaking. In the limit of very large anisotropy we find that the DC conductivity vanishes as a consequence of dimensionality reduction. We also find that a metal-insulator transition arises driven by the anisotropy.

Skill composition is a growing area of interest within reinforcement learningresearch. This approach promotes efficient use of knowledge and represents arealistic, human-like style of learning. Existing work has demonstrated how simple skills can be composed using Boolean operators to solve new, unseen taskswithout further learning. However, this approach assumes that the learned value functions for each atomic skill are optimal, an assumption which is violated in most practical cases. We propose a method that instead learns operators forcomposition using evolutionary strategies. We empirically verify our approach in tabular and high-dimensional environments. Results demonstrate that our approach outperforms existing composition methods when faced with learned,suboptimal behaviours, while also promoting robust agents and allowing for transfer between domains.

Offline multi-agent reinforcement learning (MARL) leverages static datasets of experience to learn optimal multi-agent control. However, learning from static data presents several unique challenges to overcome. In this paper, we focus on coordination failure and investigate the role of joint actions in multi-agent policy gradients with offline data, focusing on a common setting we refer to as the 'Best Response Under Data' (BRUD) approach. By using two-player polynomial games as an analytical tool, we demonstrate a simple yet overlooked failure mode of BRUD-based algorithms, which can lead to catastrophic coordination failure in the offline setting. Building on these insights, we propose an approach to mitigate such failure, by prioritising samples from the dataset based on joint-action similarity during policy learning and demonstrate its effectiveness in detailed experiments. More generally, however, we argue that prioritised dataset sampling is a promising area for innovation in offline MARL that can be combined with other effective approaches such as critic and policy regularisation. Importantly, our work shows how insights drawn from simplified, tractable games can lead to useful, theoretically grounded insights that transfer to more complex contexts. A core dimension of offering is an interactive notebook, from which almost all of our results can be reproduced, in a browser.

South Africa stands at a crucial juncture, grappling with interwoven socio-economic challenges such as poverty, inequality, unemployment, and the looming climate crisis. The government's Just Transition framework aims to enhance climate resilience, achieve net-zero greenhouse gas emissions by 2050, and promote social inclusion and poverty eradication. According to the Presidential Commission on the Fourth Industrial Revolution, artificial intelligence technologies offer significant promise in addressing these challenges. This paper explores the overlooked potential of Reinforcement Learning (RL) in supporting South Africa's Just Transition. It examines how RL can enhance agriculture and land-use practices, manage complex, decentralised energy networks, and optimise transportation and logistics, thereby playing a critical role in achieving a just and equitable transition to a low-carbon future for all South Africans. We provide a roadmap as to how other researchers in the field may be able to contribute to these pressing problems.

Being able to harness the power of large datasets for developing cooperative multi-agent controllers promises to unlock enormous value for real-world applications. Many important industrial systems are multi-agent in nature and are difficult to model using bespoke simulators. However, in industry, distributed processes can often be recorded during operation, and large quantities of demonstrative data stored. Offline multi-agent reinforcement learning (MARL) provides a promising paradigm for building effective decentralised controllers from such datasets. However, offline MARL is still in its infancy and therefore lacks standardised benchmark datasets and baselines typically found in more mature subfields of reinforcement learning (RL). These deficiencies make it difficult for the community to sensibly measure progress. In this work, we aim to fill this gap by releasing off-the-grid MARL (OG-MARL): a growing repository of high-quality datasets with baselines for cooperative offline MARL research. Our datasets provide settings that are characteristic of real-world systems, including complex environment dynamics, heterogeneous agents, non-stationarity, many agents, partial observability, suboptimality, sparse rewards and demonstrated coordination. For each setting, we provide a range of different dataset types (e.g. Good, Medium, Poor, and Replay) and profile the composition of experiences for each dataset. We hope that OG-MARL will serve the community as a reliable source of datasets and help drive progress, while also providing an accessible entry point for researchers new to the field.

Offline multi-agent reinforcement learning (MARL) is an emerging field with great promise for real-world applications. Unfortunately, the current state of research in offline MARL is plagued by inconsistencies in baselines and evaluation protocols, which ultimately makes it difficult to accurately assess progress, trust newly proposed innovations, and allow researchers to easily build upon prior work. In this paper, we firstly identify significant shortcomings in existing methodologies for measuring the performance of novel algorithms through a representative study of published offline MARL work. Secondly, by directly comparing to this prior work, we demonstrate that simple, well-implemented baselines can achieve state-of-the-art (SOTA) results across a wide range of tasks. Specifically, we show that on 35 out of 47 datasets used in prior work (almost 75% of cases), we match or surpass the performance of the current purported SOTA. Strikingly, our baselines often substantially outperform these more sophisticated algorithms. Finally, we correct for the shortcomings highlighted from this prior work by introducing a straightforward standardised methodology for evaluation and by providing our baseline implementations with statistically robust results across several scenarios, useful for comparisons in future work. Our proposal includes simple and sensible steps that are easy to adopt, which in combination with solid baselines and comparative results, could substantially improve the overall rigour of empirical science in offline MARL moving forward.

Offline multi-agent reinforcement learning (MARL) is an exciting direction of research that uses static datasets to find optimal control policies for multi-agent systems. Though the field is by definition data-driven, efforts have thus far neglected data in their drive to achieve state-of-the-art results. We first substantiate this claim by surveying the literature, showing how the majority of works generate their own datasets without consistent methodology and provide sparse information about the characteristics of these datasets. We then show why neglecting the nature of the data is problematic, through salient examples of how tightly algorithmic performance is coupled to the dataset used, necessitating a common foundation for experiments in the field. In response, we take a big step towards improving data usage and data awareness in offline MARL, with three key contributions: (1) a clear guideline for generating novel datasets; (2) a standardisation of over 80 existing datasets, hosted in a publicly available repository, using a consistent storage format and easy-to-use API; and (3) a suite of analysis tools that allow us to understand these datasets better, aiding further development.

A non-invasive technique named positron emission particle tracking (PEPT) is used to study the dynamics of mineral particles found in froth flotation. High frequency tracking is required to fully resolve turbulent flow paths with highly fluctuating velocity components. Tracking frequency is also called “location rate”, which is the number of tracked locations per second. A time interpolation algorithm was implemented on the raw data of a PEPT measurement to increase the effective location rate of PEPT data. The algorithm was validated with Monte-Carlo simulations of the Siemens ECAT “EXACT3D” HR++ PET scanner at PEPT Cape Town (University of Cape Town). The paths generated from the simulated data with improved timing information had an order of magnitude lower uncertainty in both the position and time components of a location measurement. The algorithm was then applied to PEPT experiments with freely moving particles in a laboratory scale flotation vessel to demonstrate the impact of tracking with higher fidelity.

Active Inference is a recent framework for modeling planning under uncertainty. Empirical and theoretical work have now begun to evaluate the strengths and weaknesses of this approach and how it might be improved. A recent extension - the sophisticated inference (SI) algorithm - improves performance on multi-step planning problems through recursive decision tree search. However, little work to date has been done to compare SI to other established planning algorithms. SI was also developed with a focus on inference as opposed to learning. The present paper has two aims. First, we compare performance of SI to Bayesian reinforcement learning (RL) schemes designed to solve similar problems. Second, we present an extension of SI - sophisticated learning (SL) - that more fully incorporates active learning during planning. SL maintains beliefs about how model parameters would change under the future observations expected under each policy. This allows a form of counterfactual retrospective inference in which the agent considers what could be learned from current or past observations given different future observations. To accomplish these aims, we make use of a novel, biologically inspired environment designed to highlight the problem structure for which SL offers a unique solution. Here, an agent must continually search for available (but changing) resources in the presence of competing affordances for information gain. Our simulations show that SL outperforms all other algorithms in this context - most notably, Bayes-adaptive RL and upper confidence bound algorithms, which aim to solve multi-step planning problems using similar principles (i.e., directed exploration and counterfactual reasoning). These results provide added support for the utility of Active Inference in solving this class of biologically-relevant problems and offer added tools for testing hypotheses about human cognition.

Positron emission particle tracking (PEPT) has offered important insights into the internal dynamics of multiphase flows. High precision and frequency measurements of the location of the tracer particle are required to resolve individual eddies at the millimetre scale or smaller. To explore the potential of PEPT to perform these measurements, a model was developed of the Siemens ECAT “EXACT3D” HR++ positron emission tomography (PET) scanner at the PEPT Cape Town facility in South Africa with the software Geant4 Application for Tomographic Emission (GATE) and was used to generate Lagrangian tracks from simulations of moving tracer particles. The model was validated with measurements from both experiment and simulation and was extended to two virtual scenarios inspired by turbulent flows. The location data from the simulation accurately captured linear portions of an oscillating path up to high speeds of 25 m s−1; however, tracking tended to undercut the turning points due to the high tracer acceleration. For a particle moving on a spiral path of decreasing radius, the location data tracked the path above a radius of 2.0 mm with an uncertainty equivalent to the radius of the tracer particle, 300 μm. Improvements to the measurement are required to track sub-millimetre flow structures, such as the application of PET scanners with higher spatial resolution and upgrades to the sampling processes used in location algorithms.

‘Reincarnation’ in reinforcement learning has been proposed as a formalisation of reusing prior computation from past experiments when training an agent in an environment. In this paper, we present a brief foray into the paradigm of reincarnation in the multi-agent (MA) context. We consider the case where only some agents are reincarnated, whereas the others are trained from scratch – selective reincarnation. In the fully-cooperative MA setting with heterogeneous agents, we demonstrate that selective reincarnation can lead to higher returns than training fully from scratch, and faster convergence than training with full reincarnation. However, the choice of which agents to reincarnate in a heterogeneous system is vitally important to the outcome of the training – in fact, a poor choice can lead to considerably worse results than the alternatives. We argue that a rich field of work exists here, and we hope that our effort catalyses further energy in bringing the topic of reincarnation to the multi-agent realm.

Methodologies to design efficient, affordable, and sustainable catalysts have advanced rapidly in recent years. With advances in computational power and the rapid development of computational methods, the scientific community is turning to material simulations to investigate the hidden potential of a plethora of possibly undiscovered materials in incredibly short timeframes, usually impossible via trial-and-error experimental approaches. This review article provides an overview of evolutionary-based optimization techniques with a special focus on Genetic Algorithms (GA’s) and their potential use in the catalyst design process. The ‘descriptors’ required to design catalysts via evolutionary-based optimization techniques are discussed explicitly for five key chemical reactions, namely, the Oxygen Evolution Reaction (OER), the Oxygen Reduction Reaction (ORR), the Hydrogen Evolution Reaction (HER), the Nitrogen Reduction Reaction (NRR), and the CO2 Reduction Reaction (CO2RR). The descriptors and their appraisal discussed in this review will facilitate researchers using evolutionary-based optimization techniques for catalyst design and discovery.

Text-based environments enable RL agents to learn to converse and perform interactive tasks through natural language. However, previous RL approaches applied to text-based environments show poor performance when evaluated on unseen games. This paper investigates the improvement of generalisation performance through the simple switch from a value-based update method to a policy-based one, within text-based environments. We show that by replacing commonly used value-based methods with REINFORCE with baseline, a far more general agent is produced. The policy-based agent is evaluated on Coin Collector and Question Answering with interactive text (QAit), two text-based environments designed to test zero-shot performance. We see substantial improvements on a variety of zero-shot evaluation experiments, including tripling accuracy on various QAit benchmark configurations. The results indicate that policy-based RL has significantly better generalisation capabilities than value-based methods within such text-based environments, suggesting that RL agents could be applied to more complex natural language environments.

'Reincarnation' in reinforcement learning has been proposed as a formalisation of reusing prior computation from past experiments when training an agent in an environment. In this paper, we present a brief foray into the paradigm of reincarnation in the multi-agent (MA) context. We consider the case where only some agents are reincarnated, whereas the others are trained from scratch – selective reincarnation. In the fully-cooperative MA setting with heterogeneous agents, we demonstrate that selective reincarnation can lead to higher returns than training fully from scratch, and faster convergence than training with full reincarnation. However, the choice of which agents to reincarnate in a heterogeneous system is vitally important to the outcome of the training – in fact, a poor choice can lead to considerably worse results than the alternatives. We argue that a rich field of work exists here, and we hope that our effort catalyses further energy in bringing the topic of reincarnation to the multi-agent realm.

We investigate the thermodynamics of the anisotropic magnetic AdS5 black brane solution found by D’Hoker and Kraus [1]. This solution is the gravity dual of a strongly coupled $$ \mathcal{N} $$ = 4 Super Yang-Mills plasma in ℝ3, 1, with temperature T, in the presence of a magnetic field $$ \mathcal{B} $$ . Following the procedure of holographic renormalisation we calculate the Gibbs free energy and the holographic stress tensor of the conformal plasma. We evaluate several thermodynamic quantities including the magnetisation, the anisotropic pressures and the speeds of sound. Our results are consistent with an RG flow from a perturbed AdS5 black brane at small $$ \mathcal{B} $$ /T2 to a BTZ × ℝ2 black brane at large $$ \mathcal{B} $$ /T2. We also perform a phenomenological analysis where we compare the thermodynamics of a magnetised conformal plasma against the lattice QCD results for the thermodynamics of the magnetised quark-gluon plasma.

CONTEXTDropout from engineering studies at tertiary level remains a persistent global problem.The social psychology theory of mindset explains how behaviour necessary for successful engagement with challenging academic content can be derailed by beliefs about intelligence as fixed-at-birth rather than growth mindset beliefs that intelligence can always be further developed.Given the complexity of research involving humans and the early stage of mindset research in tertiary settings, it is not surprising that the results of a recent systematic literature review on growth mindset interventions in engineering education did not identify a leading intervention.However, the review suggested that growth mindset interventions should address the broader education context and not only individual students. PURPOSE OR GOALOf all subjects, mathematics is one where fixed mindset beliefs are more frequently seen in the general population.High performing students may be at risk from the negative effects of a fixed mindset when they encounter new challenges at university.This research explores the potential of creating growth or fixed mindsets through the words used in mathematics questions.Examples from mathematics assessment tasks will be analysed to see how they align with mindset principles described in a taxonomy by Boaler (2015). APPROACH OR METHODOLOGY/METHODSA modified version of the Delphi Technique was used to reach consensus on the applicability of Boaler's taxonomy to undergraduate mathematics courses.Questions from past assessments from first-year mathematics courses were compiled, based on their potential to match the categories in Boaler's taxonomy.In six meetings over three months, all three authors discussed and classified the selected questions into the categories from Boaler's taxonomy.Where questions did not fit, modifications were brainstormed to see if modified questions could align with one or more categories from the taxonomy. ACTUAL OR ANTICIPATED OUTCOMESExamples matching all categories of Boaler's taxonomy are presented and contrasted with non-examples on the same mathematics topics. CONCLUSIONS/RECOMMENDATIONS/SUMMARYBoaler's taxonomy can guide the design of mathematics questions so that they can also reinforce growth mindset beliefs.Utilising Boaler's taxonomy in addition to the wellestablished Bloom's taxonomy to guide question setting may increase the possibility of promoting growth mindset.Multiple directions for future research are described.