Arnu Pretorius

Clément Bonnet, Daniel Luo, Donal Byrne et al.

Open-source reinforcement learning (RL) environments have played a crucial role in driving progress in the development of AI algorithms. In modern RL research, there is a need for simulated environments that are performant, scalable, and modular to enable their utilization in a wider range of potential real-world applications. Therefore, we present Jumanji, a suite of diverse RL environments specifically designed to be fast, flexible, and scalable. Jumanji provides a suite of environments focusing on combinatorial problems frequently encountered in industry, as well as challenging general decision-making tasks. By leveraging the efficiency of JAX and hardware accelerators like GPUs and TPUs, Jumanji enables rapid iteration of research ideas and large-scale experimentation, ultimately empowering more capable agents. Unlike existing RL environment suites, Jumanji is highly customizable, allowing users to tailor the initial state distribution and problem complexity to their needs. Furthermore, we provide actor-critic baselines for each environment, accompanied by preliminary findings on scaling and generalization scenarios. Jumanji aims to set a new standard for speed, adaptability, and scalability of RL environments.

Andries Smit, Paul Duckworth, Nathan Grinsztajn et al.

Recent advancements in large language models (LLMs) underscore their potential for responding to inquiries in various domains. However, ensuring that generative agents provide accurate and reliable answers remains an ongoing challenge. In this context, multi-agent debate (MAD) has emerged as a promising strategy for enhancing the truthfulness of LLMs. We benchmark a range of debating and prompting strategies to explore the trade-offs between cost, time, and accuracy. Importantly, we find that multi-agent debating systems, in their current form, do not reliably outperform other proposed prompting strategies, such as self-consistency and ensembling using multiple reasoning paths. However, when performing hyperparameter tuning, several MAD systems, such as Multi-Persona, perform better. This suggests that MAD protocols might not be inherently worse than other approaches, but that they are more sensitive to different hyperparameter settings and difficult to optimize. We build on these results to offer insights into improving debating strategies, such as adjusting agent agreement levels, which can significantly enhance performance and even surpass all other non-debate protocols we evaluated. We provide an open-source repository to the community with several state-of-the-art protocols together with evaluation scripts to benchmark across popular research datasets.

Rihab Gorsane, Omayma Mahjoub, Ruan de Kock et al.

Multi-agent reinforcement learning (MARL) has emerged as a useful approach to solving decentralised decision-making problems at scale. Research in the field has been growing steadily with many breakthrough algorithms proposed in recent years. In this work, we take a closer look at this rapid development with a focus on evaluation methodologies employed across a large body of research in cooperative MARL. By conducting a detailed meta-analysis of prior work, spanning 75 papers accepted for publication from 2016 to 2022, we bring to light worrying trends that put into question the true rate of progress. We further consider these trends in a wider context and take inspiration from single-agent RL literature on similar issues with recommendations that remain applicable to MARL. Combining these recommendations, with novel insights from our analysis, we propose a standardised performance evaluation protocol for cooperative MARL. We argue that such a standard protocol, if widely adopted, would greatly improve the validity and credibility of future research, make replication and reproducibility easier, as well as improve the ability of the field to accurately gauge the rate of progress over time by being able to make sound comparisons across different works. Finally, we release our meta-analysis data publicly on our project website for future research on evaluation: https://sites.google.com/view/marl-standard-protocol

Felix Chalumeau, Shikha Surana, Clement Bonnet et al.

Combinatorial Optimization underpins many real-world applications and yet, designing performant algorithms to solve these complex, typically NP-hard, problems remains a significant research challenge. Reinforcement Learning (RL) provides a versatile framework for designing heuristics across a broad spectrum of problem domains. However, despite notable progress, RL has not yet supplanted industrial solvers as the go-to solution. Current approaches emphasize pre-training heuristics that construct solutions but often rely on search procedures with limited variance, such as stochastically sampling numerous solutions from a single policy or employing computationally expensive fine-tuning of the policy on individual problem instances. Building on the intuition that performant search at inference time should be anticipated during pre-training, we propose COMPASS, a novel RL approach that parameterizes a distribution of diverse and specialized policies conditioned on a continuous latent space. We evaluate COMPASS across three canonical problems - Travelling Salesman, Capacitated Vehicle Routing, and Job-Shop Scheduling - and demonstrate that our search strategy (i) outperforms state-of-the-art approaches on 11 standard benchmarking tasks and (ii) generalizes better, surpassing all other approaches on a set of 18 procedurally transformed instance distributions.

Matthew Morris, Thomas D. Barrett, Arnu Pretorius

Allowing agents to share information through communication is crucial for solving complex tasks in multi-agent reinforcement learning. In this work, we consider the question of whether a given communication protocol can express an arbitrary policy. By observing that many existing protocols can be viewed as instances of graph neural networks (GNNs), we demonstrate the equivalence of joint action selection to node labelling. With standard GNN approaches provably limited in their expressive capacity, we draw from existing GNN literature and consider augmenting agent observations with: (1) unique agent IDs and (2) random noise. We provide a theoretical analysis as to how these approaches yield universally expressive communication, and also prove them capable of targeting arbitrary sets of actions for identical agents. Empirically, these augmentations are found to improve performance on tasks where expressive communication is required, whilst, in general, the optimal communication protocol is found to be task-dependent.

Callum Rhys Tilbury, Claude Formanek, Louise Beyers et al.

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.

Claude Formanek, Asad Jeewa, Jonathan Shock et al.

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.

Kale-ab Tessera, Callum Rhys Tilbury, Sasha Abramowitz et al.

Optimising deep neural networks is a challenging task due to complex training dynamics, high computational requirements, and long training times. To address this difficulty, we propose the framework of Generalisable Agents for Neural Network Optimisation (GANNO) -- a multi-agent reinforcement learning (MARL) approach that learns to improve neural network optimisation by dynamically and responsively scheduling hyperparameters during training. GANNO utilises an agent per layer that observes localised network dynamics and accordingly takes actions to adjust these dynamics at a layerwise level to collectively improve global performance. In this paper, we use GANNO to control the layerwise learning rate and show that the framework can yield useful and responsive schedules that are competitive with handcrafted heuristics. Furthermore, GANNO is shown to perform robustly across a wide variety of unseen initial conditions, and can successfully generalise to harder problems than it was trained on. Our work presents an overview of the opportunities that this paradigm offers for training neural networks, along with key challenges that remain to be overcome.

Claude Formanek, Louise Beyers, Callum Rhys Tilbury et al.

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.

Claude Formanek, Callum Rhys Tilbury, Jonathan Shock et al.

'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.

Ulrich A. Mbou Sob, Qiulin Li, Miguel Arbesú et al.

A specific challenge with deep learning approaches for molecule generation is generating both syntactically valid and chemically plausible molecular string representations. To address this, we propose a novel generative latent-variable transformer model for small molecules that leverages a recently proposed molecular string representation called SAFE. We introduce a modification to SAFE to reduce the number of invalid fragmented molecules generated during training and use this to train our model. Our experiments show that our model can generate novel molecules with a validity rate > 90% and a fragmentation rate < 1% by sampling from a latent space. By fine-tuning the model using reinforcement learning to improve molecular docking, we significantly increase the number of hit candidates for five specific protein targets compared to the pre-trained model, nearly doubling this number for certain targets. Additionally, our top 5% mean docking scores are comparable to the current state-of-the-art (SOTA), and we marginally outperform SOTA on three of the five targets.

Prince Mensah, Pelumi Victor Aderinto, Ibrahim Salihu Yusuf et al.

Accurate retrieval of vegetation biophysical variables from satellite imagery is crucial for ecosystem monitoring and agricultural management. In this work, we propose a physics-informed Transformer-VAE architecture to invert the PROSAIL radiative transfer model for simultaneous estimation of key canopy parameters from Sentinel-2 data. Unlike previous hybrid approaches that require real satellite images for self-supevised training. Our model is trained exclusively on simulated data, yet achieves performance on par with state-of-the-art methods that utilize real imagery. The Transformer-VAE incorporates the PROSAIL model as a differentiable physical decoder, ensuring that inferred latent variables correspond to physically plausible leaf and canopy properties. We demonstrate retrieval of leaf area index (LAI) and canopy chlorophyll content (CCC) on real-world field datasets (FRM4Veg and BelSAR) with accuracy comparable to models trained with real Sentinel-2 data. Our method requires no in-situ labels or calibration on real images, offering a cost-effective and self-supervised solution for global vegetation monitoring. The proposed approach illustrates how integrating physical models with advanced deep networks can improve the inversion of RTMs, opening new prospects for large-scale, physically-constrained remote sensing of vegetation traits.

Asim Osman, Sasha Abramowitz, Mark Bergh et al.

Contrastive reinforcement learning (CRL) learns goal-conditioned Q-values through a contrastive objective over state-action and goal representations, removing the need for hand-crafted reward functions. Despite impressive success in achieving viable self-supervised learning in RL, all existing CRL algorithms rely on off-policy optimisation and are mostly constrained to continuous action spaces, with little research invested in discrete environments. This leaves CRL disconnected from widely used and effective, modern on-policy training pipelines adopted across both single-agent and multi-agent RL in continuous and discrete environments. To establish a first connection, we introduce Contrastive Proximal Policy Optimisation (CPPO). CPPO is an on-policy contrastive RL algorithm that derives policy advantages directly from contrastive Q-values and optimises them via the standard PPO objective, without requiring a reward function or a replay buffer. We evaluate CPPO across continuous and discrete, single-agent and cooperative multi-agent tasks. Whilst the existence of an on-policy approach is inherently useful, we observe that \textbf{CPPO not only significantly outperforms the previous CRL baselines in 14 out of 18 tasks, but also matches or exceeds PPO's performance, which uses hand-crafted dense rewards, in 12 out of the 18 tasks tested.}

Ibrahim Salihu Yusuf, Iffanice Houndayi, Rym Oualha et al.

Open-access multispectral imagery from missions like Landsat 8-9 and Sentinel-2 has fueled the development of geospatial foundation models (GFMs) for humanitarian and environmental applications. Yet, their deployment remains limited by (i) the absence of automated geospatial data pipelines and (ii) the large size of fine-tuned models. Existing GFMs lack workflows for processing raw satellite imagery, and downstream adaptations often retain the full complexity of the original encoder. We present InstaGeo, an open-source, end-to-end framework that addresses these challenges by integrating: (1) automated data curation to transform raw imagery into model-ready datasets; (2) task-specific model distillation to derive compact, compute-efficient models; and (3) seamless deployment as interactive web-map applications. Using InstaGeo, we reproduced datasets from three published studies and trained models with marginal mIoU differences of -0.73 pp for flood mapping, -0.20 pp for crop segmentation, and +1.79 pp for desert locust prediction. The distilled models are up to 8x smaller than standard fine-tuned counterparts, reducing FLOPs and CO2 emissions with minimal accuracy loss. Leveraging InstaGeo's streamlined data pipeline, we also curated a larger crop segmentation dataset, achieving a state-of-the-art mIoU of 60.65%, a 12 pp improvement over prior baselines. Moreover, InstaGeo enables users to progress from raw data to model deployment within a single working day. By unifying data preparation, model compression, and deployment, InstaGeo transforms research-grade GFMs into practical, low-carbon tools for real-time, large-scale Earth observation. This approach shifts geospatial AI toward data quality and application-driven innovation. Source code, datasets, and model checkpoints are available at: https://github.com/instadeepai/InstaGeo-E2E-Geospatial-ML.git

Ruan de Kock, Omayma Mahjoub, Sasha Abramowitz et al.

Multi-agent reinforcement learning (MARL) research is inherently computationally expensive and it is often difficult to obtain a sufficient number of experiment samples to test hypotheses and make robust statistical claims. Furthermore, MARL algorithms are typically complex in their design and can be tricky to implement correctly. These aspects of MARL present a difficult challenge when it comes to creating useful software for advanced research. Our criteria for such software is that it should be simple enough to use to implement new ideas quickly, while at the same time be scalable and fast enough to test those ideas in a reasonable amount of time. In this preliminary technical report, we introduce Mava, a research library for MARL written purely in JAX, that aims to fulfill these criteria. We discuss the design and core features of Mava, and demonstrate its use and performance across a variety of environments. In particular, we show Mava's substantial speed advantage, with improvements of 10-100x compared to other popular MARL frameworks, while maintaining strong performance. This allows for researchers to test ideas in a few minutes instead of several hours. Finally, Mava forms part of an ecosystem of libraries that seamlessly integrate with each other to help facilitate advanced research in MARL. We hope Mava will benefit the community and help drive scientifically sound and statistically robust research in the field. The open-source repository for Mava is available at https://github.com/instadeepai/Mava.

Ibrahim Salihu Yusuf, Mukhtar Opeyemi Yusuf, Kobby Panford-Quainoo et al.

Desert locust swarms present a major threat to agriculture and food security. Addressing this challenge, our study develops an operationally-ready model for predicting locust breeding grounds, which has the potential to enhance early warning systems and targeted control measures. We curated a dataset from the United Nations Food and Agriculture Organization's (UN-FAO) locust observation records and analyzed it using two types of spatio-temporal input features: remotely-sensed environmental and climate data as well as multi-spectral earth observation images. Our approach employed custom deep learning models (three-dimensional and LSTM-based recurrent convolutional networks), along with the geospatial foundational model Prithvi recently released by Jakubik et al., 2023. These models notably outperformed existing baselines, with the Prithvi-based model, fine-tuned on multi-spectral images from NASA's Harmonized Landsat and Sentinel-2 (HLS) dataset, achieving the highest accuracy, F1 and ROC-AUC scores (83.03%, 81.53% and 87.69%, respectively). A significant finding from our research is that multi-spectral earth observation images alone are sufficient for effective locust breeding ground prediction without the need to explicitly incorporate climatic or environmental features.

Omayma Mahjoub, Ruan de Kock, Siddarth Singh et al.

Measuring the contribution of individual agents is challenging in cooperative multi-agent reinforcement learning (MARL). In cooperative MARL, team performance is typically inferred from a single shared global reward. Arguably, among the best current approaches to effectively measure individual agent contributions is to use Shapley values. However, calculating these values is expensive as the computational complexity grows exponentially with respect to the number of agents. In this paper, we adapt difference rewards into an efficient method for quantifying the contribution of individual agents, referred to as Agent Importance, offering a linear computational complexity relative to the number of agents. We show empirically that the computed values are strongly correlated with the true Shapley values, as well as the true underlying individual agent rewards, used as the ground truth in environments where these are available. We demonstrate how Agent Importance can be used to help study MARL systems by diagnosing algorithmic failures discovered in prior MARL benchmarking work. Our analysis illustrates Agent Importance as a valuable explainability component for future MARL benchmarks.

Siddarth Singh, Omayma Mahjoub, Ruan de Kock et al.

Establishing sound experimental standards and rigour is important in any growing field of research. Deep Multi-Agent Reinforcement Learning (MARL) is one such nascent field. Although exciting progress has been made, MARL has recently come under scrutiny for replicability issues and a lack of standardised evaluation methodology, specifically in the cooperative setting. Although protocols have been proposed to help alleviate the issue, it remains important to actively monitor the health of the field. In this work, we extend the database of evaluation methodology previously published by containing meta-data on MARL publications from top-rated conferences and compare the findings extracted from this updated database to the trends identified in their work. Our analysis shows that many of the worrying trends in performance reporting remain. This includes the omission of uncertainty quantification, not reporting all relevant evaluation details and a narrowing of algorithmic development classes. Promisingly, we do observe a trend towards more difficult scenarios in SMAC-v1, which if continued into SMAC-v2 will encourage novel algorithmic development. Our data indicate that replicability needs to be approached more proactively by the MARL community to ensure trust in the field as we move towards exciting new frontiers.

Jemma Daniel, Ruan de Kock, Louay Ben Nessir et al.

The Transformer model has demonstrated success across a wide range of domains, including in Multi-Agent Reinforcement Learning (MARL) where the Multi-Agent Transformer (MAT) has emerged as a leading algorithm in the field. However, a significant drawback of Transformer models is their quadratic computational complexity relative to input size, making them computationally expensive when scaling to larger inputs. This limitation restricts MAT's scalability in environments with many agents. Recently, State-Space Models (SSMs) have gained attention due to their computational efficiency, but their application in MARL remains unexplored. In this work, we investigate the use of Mamba, a recent SSM, in MARL and assess whether it can match the performance of MAT while providing significant improvements in efficiency. We introduce a modified version of MAT that incorporates standard and bi-directional Mamba blocks, as well as a novel "cross-attention" Mamba block. Extensive testing shows that our Multi-Agent Mamba (MAM) matches the performance of MAT across multiple standard multi-agent environments, while offering superior scalability to larger agent scenarios. This is significant for the MARL community, because it indicates that SSMs could replace Transformers without compromising performance, whilst also supporting more effective scaling to higher numbers of agents. Our project page is available at https://sites.google.com/view/multi-agent-mamba .

Wiem Khlifi, Siddarth Singh, Omayma Mahjoub et al.

Cooperative multi-agent reinforcement learning (MARL) has made substantial strides in addressing the distributed decision-making challenges. However, as multi-agent systems grow in complexity, gaining a comprehensive understanding of their behaviour becomes increasingly challenging. Conventionally, tracking team rewards over time has served as a pragmatic measure to gauge the effectiveness of agents in learning optimal policies. Nevertheless, we argue that relying solely on the empirical returns may obscure crucial insights into agent behaviour. In this paper, we explore the application of explainable AI (XAI) tools to gain profound insights into agent behaviour. We employ these diagnostics tools within the context of Level-Based Foraging and Multi-Robot Warehouse environments and apply them to a diverse array of MARL algorithms. We demonstrate how our diagnostics can enhance the interpretability and explainability of MARL systems, providing a better understanding of agent behaviour.

Claude Formanek, Omayma Mahjoub, Louay Ben Nessir et al.

A key challenge in offline multi-agent reinforcement learning (MARL) is achieving effective many-agent multi-step coordination in complex environments. In this work, we propose Oryx, a novel algorithm for offline cooperative MARL to directly address this challenge. Oryx adapts the recently proposed retention-based architecture Sable and combines it with a sequential form of implicit constraint Q-learning (ICQ), to develop a novel offline autoregressive policy update scheme. This allows Oryx to solve complex coordination challenges while maintaining temporal coherence over long trajectories. We evaluate Oryx across a diverse set of benchmarks from prior works -- SMAC, RWARE, and Multi-Agent MuJoCo -- covering tasks of both discrete and continuous control, varying in scale and difficulty. Oryx achieves state-of-the-art performance on more than 80% of the 65 tested datasets, outperforming prior offline MARL methods and demonstrating robust generalisation across domains with many agents and long horizons. Finally, we introduce new datasets to push the limits of many-agent coordination in offline MARL, and demonstrate Oryx's superior ability to scale effectively in such settings.

Felix Chalumeau, Daniel Rajaonarivonivelomanantsoa, Ruan de Kock et al.

Reinforcement learning (RL) systems have countless applications, from energy-grid management to protein design. However, such real-world scenarios are often extremely difficult, combinatorial in nature, and require complex coordination between multiple agents. This level of complexity can cause even state-of-the-art RL systems, trained until convergence, to hit a performance ceiling which they are unable to break out of with zero-shot inference. Meanwhile, many digital or simulation-based applications allow for an inference phase that utilises a specific time and compute budget to explore multiple attempts before outputting a final solution. In this work, we show that such an inference phase employed at execution time, and the choice of a corresponding inference strategy, are key to breaking the performance ceiling observed in complex multi-agent RL problems. Our main result is striking: we can obtain up to a 126% and, on average, a 45% improvement over the previous state-of-the-art across 17 tasks, using only a couple seconds of extra wall-clock time during execution. We also demonstrate promising compute scaling properties, supported by over 60k experiments, making it the largest study on inference strategies for complex RL to date. Our experimental data and code are available at https://sites.google.com/view/inference-strategies-rl.

Felix Chalumeau, Refiloe Shabe, Noah De Nicola et al.

Routing Problems are central to many real-world applications, yet remain challenging due to their (NP-)hard nature. Amongst existing approaches, heuristics often offer the best trade-off between quality and scalability, making them suitable for industrial use. While Reinforcement Learning (RL) offers a flexible framework for designing heuristics, its adoption over handcrafted heuristics remains incomplete. Existing learned methods still lack the ability to adapt to specific instances and fully leverage the available computational budget. Current best methods either rely on a collection of pre-trained policies, or on RL fine-tuning; hence failing to fully utilize newly available information within the constraints of the budget. In response, we present MEMENTO, an approach that leverages memory to improve the search of neural solvers at inference. MEMENTO leverages online data collected across repeated attempts to dynamically adjust the action distribution based on the outcome of previous decisions. We validate its effectiveness on the Traveling Salesman and Capacitated Vehicle Routing problems, demonstrating its superiority over tree-search and policy-gradient fine-tuning; and showing that it can be zero-shot combined with diversity-based solvers. We successfully train all RL auto-regressive solvers on large instances, and verify MEMENTO's scalability and data-efficiency: pushing the state-of-the-art on 11 out of 12 evaluated tasks.

Claude Formanek, Callum Rhys Tilbury, Louise Beyers et al.

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.

St John Grimbly, Jonathan Shock, Arnu Pretorius

This paper serves to introduce the reader to the field of multi-agent reinforcement learning (MARL) and its intersection with methods from the study of causality. We highlight key challenges in MARL and discuss these in the context of how causal methods may assist in tackling them. We promote moving toward a 'causality first' perspective on MARL. Specifically, we argue that causality can offer improved safety, interpretability, and robustness, while also providing strong theoretical guarantees for emergent behaviour. We discuss potential solutions for common challenges, and use this context to motivate future research directions.

Ibrahim Salihu Yusuf, Kale-ab Tessera, Thomas Tumiel et al.

Desert locust outbreaks threaten the food security of a large part of Africa and have affected the livelihoods of millions of people over the years. Machine learning (ML) has been demonstrated as an effective approach to locust distribution modelling which could assist in early warning. ML requires a significant amount of labelled data to train. Most publicly available labelled data on locusts are presence-only data, where only the sightings of locusts being present at a location are recorded. Therefore, prior work using ML have resorted to pseudo-absence generation methods as a way to circumvent this issue. The most commonly used approach is to randomly sample points in a region of interest while ensuring that these sampled pseudo-absence points are at least a specific distance away from true presence points. In this paper, we compare this random sampling approach to more advanced pseudo-absence generation methods, such as environmental profiling and optimal background extent limitation, specifically for predicting desert locust breeding grounds in Africa. Interestingly, we find that for the algorithms we tested, namely logistic regression, gradient boosting, random forests and maximum entropy, all popular in prior work, the logistic model performed significantly better than the more sophisticated ensemble methods, both in terms of prediction accuracy and F1 score. Although background extent limitation combined with random sampling boosted performance for ensemble methods, for LR this was not the case, and instead, a significant improvement was obtained when using environmental profiling. In light of this, we conclude that a simpler ML approach such as logistic regression combined with more advanced pseudo-absence generation, specifically environmental profiling, can be a sensible and effective approach to predicting locust breeding grounds across Africa.

Kevin Eloff, Okko Räsänen, Herman A. Engelbrecht et al.

Multi-agent reinforcement learning has been used as an effective means to study emergent communication between agents, yet little focus has been given to continuous acoustic communication. This would be more akin to human language acquisition; human infants acquire language in large part through continuous signalling with their caregivers. We therefore ask: Are we able to observe emergent language between agents with a continuous communication channel? Our goal is to provide a platform to begin bridging the gap between human and agent communication, allowing us to analyse continuous signals, how they emerge, their characteristics, and how they relate to human language acquisition. We propose a messaging environment where a Speaker agent needs to convey a set of attributes to a Listener over a noisy acoustic channel. Using DQN to train our agents, we show that: (1) unlike the discrete case, the acoustic Speaker learns redundancy to improve Listener coherency, (2) the acoustic Speaker develops more compositional communication protocols which implicitly compensates for transmission errors over a noisy channel, and (3) DQN has significant performance gains and increased compositionality when compared to previous methods optimised using REINFORCE.

Scott Cameron, Tyron Cameron, Arnu Pretorius et al.

Stochastic differential equations provide a rich class of flexible generative models, capable of describing a wide range of spatio-temporal processes. A host of recent work looks to learn data-representing SDEs, using neural networks and other flexible function approximators. Despite these advances, learning remains computationally expensive due to the sequential nature of SDE integrators. In this work, we propose an importance-sampling estimator for probabilities of observations of SDEs for the purposes of learning. Crucially, the approach we suggest does not rely on such integrators. The proposed method produces lower-variance gradient estimates compared to algorithms based on SDE integrators and has the added advantage of being embarrassingly parallelizable. This facilitates the effective use of large-scale parallel hardware for massive decreases in computation time.

Arnu Pretorius, Scott Cameron, Elan van Biljon et al.

Multi-agent reinforcement learning has recently shown great promise as an approach to networked system control. Arguably, one of the most difficult and important tasks for which large scale networked system control is applicable is common-pool resource management. Crucial common-pool resources include arable land, fresh water, wetlands, wildlife, fish stock, forests and the atmosphere, of which proper management is related to some of society's greatest challenges such as food security, inequality and climate change. Here we take inspiration from a recent research program investigating the game-theoretic incentives of humans in social dilemma situations such as the well-known tragedy of the commons. However, instead of focusing on biologically evolved human-like agents, our concern is rather to better understand the learning and operating behaviour of engineered networked systems comprising general-purpose reinforcement learning agents, subject only to nonbiological constraints such as memory, computation and communication bandwidth. Harnessing tools from empirical game-theoretic analysis, we analyse the differences in resulting solution concepts that stem from employing different information structures in the design of networked multi-agent systems. These information structures pertain to the type of information shared between agents as well as the employed communication protocol and network topology. Our analysis contributes new insights into the consequences associated with certain design choices and provides an additional dimension of comparison between systems beyond efficiency, robustness, scalability and mean control performance.

Elan van Biljon, Arnu Pretorius, Julia Kreutzer

Transformers have shown great promise as an approach to Neural Machine Translation (NMT) for low-resource languages. However, at the same time, transformer models remain difficult to optimize and require careful tuning of hyper-parameters to be useful in this setting. Many NMT toolkits come with a set of default hyper-parameters, which researchers and practitioners often adopt for the sake of convenience and avoiding tuning. These configurations, however, have been optimized for large-scale machine translation data sets with several millions of parallel sentences for European languages like English and French. In this work, we find that the current trend in the field to use very large models is detrimental for low-resource languages, since it makes training more difficult and hurts overall performance, confirming previous observations. We see our work as complementary to the Masakhane project ("Masakhane" means "We Build Together" in isiZulu.) In this spirit, low-resource NMT systems are now being built by the community who needs them the most. However, many in the community still have very limited access to the type of computational resources required for building extremely large models promoted by industrial research. Therefore, by showing that transformer models perform well (and often best) at low-to-moderate depth, we hope to convince fellow researchers to devote less computational resources, as well as time, to exploring overly large models during the development of these systems.

Felix McGregor, Arnu Pretorius, Johan du Preez et al.

Bayesian neural networks (BNNs) have developed into useful tools for probabilistic modelling due to recent advances in variational inference enabling large scale BNNs. However, BNNs remain brittle and hard to train, especially: (1) when using deep architectures consisting of many hidden layers and (2) in situations with large weight variances. We use signal propagation theory to quantify these challenges and propose self-stabilising priors. This is achieved by a reformulation of the ELBO to allow the prior to influence network signal propagation. Then, we develop a stabilising prior, where the distributional parameters of the prior are adjusted before each forward pass to ensure stability of the propagating signal. This stabilised signal propagation leads to improved convergence and robustness making it possible to train deeper networks and in more noisy settings.

Arnu Pretorius, Elan van Biljon, Benjamin van Niekerk et al.

Recent work in signal propagation theory has shown that dropout limits the depth to which information can propagate through a neural network. In this paper, we investigate the effect of initialisation on training speed and generalisation for ReLU networks within this depth limit. We ask the following research question: given that critical initialisation is crucial for training at large depth, if dropout limits the depth at which networks are trainable, does initialising critically still matter? We conduct a large-scale controlled experiment, and perform a statistical analysis of over $12000$ trained networks. We find that (1) trainable networks show no statistically significant difference in performance over a wide range of non-critical initialisations; (2) for initialisations that show a statistically significant difference, the net effect on performance is small; (3) only extreme initialisations (very small or very large) perform worse than criticality. These findings also apply to standard ReLU networks of moderate depth as a special case of zero dropout. Our results therefore suggest that, in the shallow-to-moderate depth setting, critical initialisation provides zero performance gains when compared to off-critical initialisations and that searching for off-critical initialisations that might improve training speed or generalisation, is likely to be a fruitless endeavour.

Arnu Pretorius, Herman Kamper, Steve Kroon

Recent work has established the equivalence between deep neural networks and Gaussian processes (GPs), resulting in so-called neural network Gaussian processes (NNGPs). The behaviour of these models depends on the initialisation of the corresponding network. In this work, we consider the impact of noise regularisation (e.g. dropout) on NNGPs, and relate their behaviour to signal propagation theory in noise regularised deep neural networks. For ReLU activations, we find that the best performing NNGPs have kernel parameters that correspond to a recently proposed initialisation scheme for noise regularised ReLU networks. In addition, we show how the noise influences the covariance matrix of the NNGP, producing a stronger prior towards simple functions away from the training points. We verify our theoretical findings with experiments on MNIST and CIFAR-10 as well as on synthetic data.

Ryan Eloff, André Nortje, Benjamin van Niekerk et al.

For our submission to the ZeroSpeech 2019 challenge, we apply discrete latent-variable neural networks to unlabelled speech and use the discovered units for speech synthesis. Unsupervised discrete subword modelling could be useful for studies of phonetic category learning in infants or in low-resource speech technology requiring symbolic input. We use an autoencoder (AE) architecture with intermediate discretisation. We decouple acoustic unit discovery from speaker modelling by conditioning the AE's decoder on the training speaker identity. At test time, unit discovery is performed on speech from an unseen speaker, followed by unit decoding conditioned on a known target speaker to obtain reconstructed filterbanks. This output is fed to a neural vocoder to synthesise speech in the target speaker's voice. For discretisation, categorical variational autoencoders (CatVAEs), vector-quantised VAEs (VQ-VAEs) and straight-through estimation are compared at different compression levels on two languages. Our final model uses convolutional encoding, VQ-VAE discretisation, deconvolutional decoding and an FFTNet vocoder. We show that decoupled speaker conditioning intrinsically improves discrete acoustic representations, yielding competitive synthesis quality compared to the challenge baseline.

Arnu Pretorius, Elan Van Biljon, Steve Kroon et al.

Stochastic regularisation is an important weapon in the arsenal of a deep learning practitioner. However, despite recent theoretical advances, our understanding of how noise influences signal propagation in deep neural networks remains limited. By extending recent work based on mean field theory, we develop a new framework for signal propagation in stochastic regularised neural networks. Our noisy signal propagation theory can incorporate several common noise distributions, including additive and multiplicative Gaussian noise as well as dropout. We use this framework to investigate initialisation strategies for noisy ReLU networks. We show that no critical initialisation strategy exists using additive noise, with signal propagation exploding regardless of the selected noise distribution. For multiplicative noise (e.g. dropout), we identify alternative critical initialisation strategies that depend on the second moment of the noise distribution. Simulations and experiments on real-world data confirm that our proposed initialisation is able to stably propagate signals in deep networks, while using an initialisation disregarding noise fails to do so. Furthermore, we analyse correlation dynamics between inputs. Stronger noise regularisation is shown to reduce the depth to which discriminatory information about the inputs to a noisy ReLU network is able to propagate, even when initialised at criticality. We support our theoretical predictions for these trainable depths with simulations, as well as with experiments on MNIST and CIFAR-10

Arnu Pretorius, Steve Kroon, Herman Kamper

Denoising autoencoders (DAEs) have proven useful for unsupervised representation learning, but a thorough theoretical understanding is still lacking of how the input noise influences learning. Here we develop theory for how noise influences learning in DAEs. By focusing on linear DAEs, we are able to derive analytic expressions that exactly describe their learning dynamics. We verify our theoretical predictions with simulations as well as experiments on MNIST and CIFAR-10. The theory illustrates how, when tuned correctly, noise allows DAEs to ignore low variance directions in the inputs while learning to reconstruct them. Furthermore, in a comparison of the learning dynamics of DAEs to standard regularised autoencoders, we show that noise has a similar regularisation effect to weight decay, but with faster training dynamics. We also show that our theoretical predictions approximate learning dynamics on real-world data and qualitatively match observed dynamics in nonlinear DAEs.