Search

Predicting the biophysical and functional properties of proteins is essential for in silico protein design. Machine learning has emerged as a promising technique for such prediction tasks. However, the relative scarcity of in vitro annotations means that these models often have little, or no, specific data on the desired fitness prediction task. As a result of limited data, protein language models (PLMs) are typically trained on general protein sequence modeling tasks, and then fine-tuned, or applied zero-shot, to protein fitness prediction. When no task data is available, the models make strong assumptions about the correlation between the protein sequence likelihood and fitness scores. In contrast, we propose meta-learning over a distribution of standard fitness prediction tasks, and demonstrate positive transfer to unseen fitness prediction tasks. Our method, called Metalic (Meta-Learning In-Context), uses in-context learning and fine-tuning, when data is available, to adapt to new tasks. Crucially, fine-tuning enables considerable generalization, even though it is not accounted for during meta-training. Our fine-tuned models achieve strong results with 18 times fewer parameters than state-of-the-art models. Moreover, our method sets a new state-of-the-art in low-data settings on ProteinGym, an established fitness-prediction benchmark. Due to data scarcity, we believe meta-learning will play a pivotal role in advancing protein engineering.

Combinatorial Optimization is crucial to numerous real-world applications, yet still presents challenges due to its (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 within industrial solvers. Existing learned methods still lack the ability to adapt to specific instances and fully leverage the available computational budget. The current best methods either rely on a collection of pre-trained policies, or on data-inefficient 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 adaptation of neural solvers at inference time. MEMENTO enables updating the action distribution dynamically based on the outcome of previous decisions. We validate its effectiveness on benchmark problems, in particular Traveling Salesman and Capacitated Vehicle Routing, demonstrating its superiority over tree-search and policy-gradient fine-tuning; and showing it can be zero-shot combined with diversity-based solvers. We successfully train all RL auto-regressive solvers on large instances, and show that MEMENTO can scale and is data-efficient. Overall, MEMENTO enables to push the state-of-the-art on 11 out of 12 evaluated tasks.

Despite being pivotal to the habitability of our planet, the process by which Earth gained its present-day hydrogen budget is unclear. Due to their isotopic similarity to terrestrial rocks across a range of elements, enstatite chondrites (ECs) are thought to be the meteorites that best represent Earth's building blocks. Because of ECs' nominally anhydrous mineralogy, these building blocks have long been presumed to have supplied negligible hydrogen to the proto-Earth. Instead, hydrogen has been proposed to have been delivered to our planet after its main stage of formation by impacts from hydrated asteroids. In this case, our planet's habitability would have its origins in a stochastic process. However, ECs have recently been found to unexpectedly contain enough hydrogen to readily explain Earth's present-day water budget. Although this result would transform the processes we believe are required for rocky planets to be suitable to life, the mineralogical source of ~80% of hydrogen in these meteorites was previously unknown. As such, the reason ECs are seemingly rich in hydrogen was unclear. Here, we apply sulfur X-ray absorption near edge structure (S-XANES) spectroscopy to ECs, finding that most (~70%) of their hydrogen is bonded to sulfur. Moreover, the concentration of the S-H bond is intimately linked to the abundance of micrometre-scale pyrrhotite (Fe1-xS, 0

Representation learning and \emph{de novo} generation of proteins are pivotal computational biology tasks. Whilst natural language processing (NLP) techniques have proven highly effective for protein sequence modelling, structure modelling presents a complex challenge, primarily due to its continuous and three-dimensional nature. Motivated by this discrepancy, we introduce an approach using a vector-quantized autoencoder that effectively tokenizes protein structures into discrete representations. This method transforms the continuous, complex space of protein structures into a manageable, discrete format with a codebook ranging from 4096 to 64000 tokens, achieving high-fidelity reconstructions with backbone root mean square deviations (RMSD) of approximately 1-5 \AA. To demonstrate the efficacy of our learned representations, we show that a simple GPT model trained on our codebooks can generate novel, diverse, and designable protein structures. Our approach not only provides representations of protein structure, but also mitigates the challenges of disparate modal representations and sets a foundation for seamless, multi-modal integration, enhancing the capabilities of computational methods in protein design.

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., Comment: 2 pages, 13 figures

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., Comment: Fix typo in formula and add a reference

Photonic qubits play an instrumental role in the development of advanced quantum technologies, including quantum networking, boson sampling and measurement based quantum computing. A promising framework for the deterministic production of indistinguishable single photons is an atomic emitter coupled to a single mode of a high finesse optical cavity. Polarisation control is an important cornerstone, particularly when the polarisation defines the state of a quantum bit. Here, we propose a scheme for producing bursts of polarised single photons by coupling a generalised atomic emitter to an optical cavity, exploiting a particular choice of quantisation axis. In connection with two re-preparation methods, simulations predict 10-photon bursts coincidence count rates on the order of 1 kHz with single 87Rb atoms trapped in a state of the art optical cavity. This paves the way for novel n-photon experiments with atom-cavity sources.

The measurement of the bandgap in quasi-one dimensional materials such as carbon nanotubes is challenging due to its dimensionality. In this work, we measure the electronic bandgap of networks of polymer-wrapped semiconducting single-walled carbon nanotubes (s-SWCNTs) using non-ideal p-n diodes. Using these diodes, we measure the electronic bandgap and excitonic levels of different polymer-wrapped s-SWCNTs with varying diameters: arc discharge (~1.55nm), (7,5) (0.83nm), and (6,5) (0.76nm). Our values are consistent with theoretical predictions, providing insight into the fundamental properties of networks of s-SWCNTs.

Photovoltaics has two main processes: Optical absorption and power conversion. In photosynthesis, the two equivalent processes are optical absorption and chemical conversion. Whereas in the latter, the two processes are carried out by distinct proteins, in conventional photovoltaic diodes, the two processes are convoluted because the optical and transport paths are the same, leading to inefficiencies. Here, we separate the site and direction of light absorption from those of power generation to show that semiconducting single-walled carbon nanotubes (s-SWCNTs) provide an artificial system that models photosynthesis in a tandem geometry. Using different s-SWCNT chiralities, we implement an energy funnel in dual-gated p-n diodes. This enables the capture of photons from multiple regions of the solar spectrum and the funneling of photogenerated excitons to the smallest bandgap s-SWCNT layer, where they become free carriers. As a result, we demonstrate an increase in the magnitude and spectral response of photocurrent by adding more s-SWCNT layers of different bandgaps without a corresponding deleterious increase in the dark leakage current.

Applying reinforcement learning (RL) to combinatorial optimization problems is attractive as it removes the need for expert knowledge or pre-solved instances. However, it is unrealistic to expect an agent to solve these (often NP-)hard problems in a single shot at inference due to their inherent complexity. Thus, leading approaches often implement additional search strategies, from stochastic sampling and beam search to explicit fine-tuning. In this paper, we argue for the benefits of learning a population of complementary policies, which can be simultaneously rolled out at inference. To this end, we introduce Poppy, a simple training procedure for populations. Instead of relying on a predefined or hand-crafted notion of diversity, Poppy induces an unsupervised specialization targeted solely at maximizing the performance of the population. We show that Poppy produces a set of complementary policies, and obtains state-of-the-art RL results on four popular NP-hard problems: traveling salesman, capacitated vehicle routing, 0-1 knapsack, and job-shop scheduling.

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., Comment: Published in NeurIPS 2022

Combinatorial optimisation problems framed as mixed integer linear programmes (MILPs) are ubiquitous across a range of real-world applications. The canonical branch-and-bound algorithm seeks to exactly solve MILPs by constructing a search tree of increasingly constrained sub-problems. In practice, its solving time performance is dependent on heuristics, such as the choice of the next variable to constrain ('branching'). Recently, machine learning (ML) has emerged as a promising paradigm for branching. However, prior works have struggled to apply reinforcement learning (RL), citing sparse rewards, difficult exploration, and partial observability as significant challenges. Instead, leading ML methodologies resort to approximating high quality handcrafted heuristics with imitation learning (IL), which precludes the discovery of novel policies and requires expensive data labelling. In this work, we propose retro branching; a simple yet effective approach to RL for branching. By retrospectively deconstructing the search tree into multiple paths each contained within a sub-tree, we enable the agent to learn from shorter trajectories with more predictable next states. In experiments on four combinatorial tasks, our approach enables learning-to-branch without any expert guidance or pre-training. We outperform the current state-of-the-art RL branching algorithm by 3-5x and come within 20% of the best IL method's performance on MILPs with 500 constraints and 1000 variables, with ablations verifying that our retrospectively constructed trajectories are essential to achieving these results., Comment: Accepted to AAAI'23: Proceedings of the Thirty-Seventh AAAI Conference on Artificial Intelligence

From logistics to the natural sciences, combinatorial optimisation on graphs underpins numerous real-world applications. Reinforcement learning (RL) has shown particular promise in this setting as it can adapt to specific problem structures and does not require pre-solved instances for these, often NP-hard, problems. However, state-of-the-art (SOTA) approaches typically suffer from severe scalability issues, primarily due to their reliance on expensive graph neural networks (GNNs) at each decision step. We introduce ECORD; a novel RL algorithm that alleviates this expense by restricting the GNN to a single pre-processing step, before entering a fast-acting exploratory phase directed by a recurrent unit. Experimentally, ECORD achieves a new SOTA for RL algorithms on the Maximum Cut problem, whilst also providing orders of magnitude improvement in speed and scalability. Compared to the nearest competitor, ECORD reduces the optimality gap by up to 73% on 500 vertex graphs with a decreased wall-clock time. Moreover, ECORD retains strong performance when generalising to larger graphs with up to 10000 vertices.

Wilhelm (2021) has recently defended a criterion for comparing structure of mathematical objects, which he calls Subgroup. He argues that Subgroup is better than SYM * , another widely adopted criterion. We argue that this is mistaken; Subgroup is strictly worse than SYM *. We then formulate a new criterion that improves on both SYM * and Subgroup, answering Wilhelm's criticisms of SYM * along the way. We conclude by arguing that no criterion that looks only to the automorphisms of mathematical objects to compare their structure can be fully satisfactory., Comment: 13 pages, 1 figure (depicting a giraffe)

We experimentally study a gas of quantum degenerate $^{87}$Rb atoms throughout the full dimensional crossover, from a one-dimensional (1D) system exhibiting phase fluctuations consistent with 1D theory to a three-dimensional (3D) phase-coherent system, thereby smoothly interpolating between these distinct, well-understood regimes. Using a hybrid trapping architecture combining an atom chip with a printed circuit board, we continuously adjust the system's dimensionality over a wide range while measuring the phase fluctuations through the power spectrum of density ripples in time-of-flight expansion. Our measurements confirm that the chemical potential $\mu$ controls the departure of the system from 3D and that the fluctuations are dependent on both $\mu$ and the temperature $T$. Through a rigorous study we quantitatively observe how inside the crossover the dependence on $T$ gradually disappears as the system becomes 3D. Throughout the entire crossover the fluctuations are shown to be determined by the relative occupation of 1D axial collective excitations., Comment: 10 pages, 6 figures

Self-tuning algorithms that adapt the learning process online encourage more effective and robust learning. Among all the methods available, meta-gradients have emerged as a promising approach. They leverage the differentiability of the learning rule with respect to some hyper-parameters to adapt them in an online fashion. Although meta-gradients can be accumulated over multiple learning steps to avoid myopic updates, this is rarely used in practice. In this work, we demonstrate that whilst multi-step meta-gradients do provide a better learning signal in expectation, this comes at the cost of a significant increase in variance, hindering performance. In the light of this analysis, we introduce a novel method mixing multiple inner steps that enjoys a more accurate and robust meta-gradient signal, essentially trading off bias and variance in meta-gradient estimation. When applied to the Snake game, the mixing meta-gradient algorithm can cut the variance by a factor of 3 while achieving similar or higher performance., Comment: 14 pages, 6 figures, 2 tables

In recent years, neural network quantum states (NNQS) have emerged as powerful tools for the study of quantum many-body systems. Electronic structure calculations are one such canonical many-body problem that have attracted significant research efforts spanning multiple decades, whilst only recently being attempted with NNQS. However, the complex non-local interactions and high sample complexity are significant challenges that call for bespoke solutions. Here, we parameterise the electronic wavefunction with a novel autoregressive neural network (ARN) that permits highly efficient and scalable sampling, whilst also embedding physical priors reflecting the structure of molecular systems without sacrificing expressibility. This allows us to perform electronic structure calculations on molecules with up to 30 spin-orbitals -- at least an order of magnitude more Slater determinants than previous applications of conventional NNQS -- and we find that our ansatz can outperform the de-facto gold-standard coupled cluster methods even in the presence of strong quantum correlations. With a highly expressive neural network for which sampling is no longer a computational bottleneck, we conclude that the barriers to further scaling are not associated with the wavefunction ansatz itself, but rather are inherent to any variational Monte Carlo approach., Comment: 8 pages, plus Methods and Supplementary Information

In his 1935 Gedankenexperiment, Erwin Schr\"{o}dinger imagined a poisonous substance which has a 50% probability of being released, based on the decay of a radioactive atom. As such, the life of the cat and the state of the poison become entangled, and the fate of the cat is determined upon opening the box. We present an experimental technique that keeps the cat alive on any account. This method relies on the time-resolved Hong-Ou-Mandel effect: two long, identical photons impinging on a beam splitter always bunch in either of the outputs. Interpreting the first photon detection as the state of the poison, the second photon is identified as the state of the cat. Even after the collapse of the first photon's state, we show their fates are intertwined through quantum interference. We demonstrate this by a sudden phase change between the inputs, administered conditionally on the outcome of the first detection, which steers the second photon to a pre-defined output and ensures that the cat is always observed alive., Comment: 7 pages, 4 figures, including supplementary material with 4 pages and 4 figures

Understanding the origin and evolution of the lunar volatile system is not only compelling lunar science, but also fundamental Solar System science. This white paper (submitted to the US National Academies' Decadal Survey in Planetary Science and Astrobiology 2023-2032) summarizes recent advances in our understanding of lunar volatiles, identifies outstanding questions for the next decade, and discusses key steps required to address these questions.

Optics is a promising platform in which to help realise the next generation of fast, parallel and energy-efficient computation. We demonstrate a reconfigurable free-space optical multiplier that is capable of over 3000 computations in parallel, using spatial light modulators with a pixel resolution of only 340x340. This enables vector-matrix multiplication and parallel vector-vector multiplication with vector size of up to 56. Our design is the first to simultaneously support optical implementation of reconfigurable, large-size and real-valued linear algebraic operations. Such an optical multiplier can serve as a building block of special-purpose optical processors such as optical neural networks and optical Ising machines., Comment: 4 pages, 4 figures

Learning disentangled representations is a key step towards effectively discovering and modelling the underlying structure of environments. In the natural sciences, physics has found great success by describing the universe in terms of symmetry preserving transformations. Inspired by this formalism, we propose a framework, built upon the theory of group representation, for learning representations of a dynamical environment structured around the transformations that generate its evolution. Experimentally, we learn the structure of explicitly symmetric environments without supervision from observational data generated by sequential interactions. We further introduce an intuitive disentanglement regularisation to ensure the interpretability of the learnt representations. We show that our method enables accurate long-horizon predictions, and demonstrate a correlation between the quality of predictions and disentanglement in the latent space., Comment: Accepted to NeurIPS 2020

Backpropagation through nonlinear neurons is an outstanding challenge to the field of optical neural networks and the major conceptual barrier to all-optical training schemes. Each neuron is required to exhibit a directionally dependent response to propagating optical signals, with the backwards response conditioned on the forward signal, which is highly non-trivial to implement optically. We propose a practical and surprisingly simple solution that uses saturable absorption to provide the network nonlinearity. We find that the backward propagating gradients required to train the network can be approximated in a pump-probe scheme that requires only passive optical elements. Simulations show that, with readily obtainable optical depths, our approach can achieve equivalent performance to state-of-the-art computational networks on image classification benchmarks, even in deep networks with multiple sequential gradient approximations. This scheme is compatible with leading optical neural network proposals and therefore provides a feasible path towards end-to-end optical training., Comment: Error fixed in Fig.1

Many real-world problems can be reduced to combinatorial optimization on a graph, where the subset or ordering of vertices that maximize some objective function must be found. With such tasks often NP-hard and analytically intractable, reinforcement learning (RL) has shown promise as a framework with which efficient heuristic methods to tackle these problems can be learned. Previous works construct the solution subset incrementally, adding one element at a time, however, the irreversible nature of this approach prevents the agent from revising its earlier decisions, which may be necessary given the complexity of the optimization task. We instead propose that the agent should seek to continuously improve the solution by learning to explore at test time. Our approach of exploratory combinatorial optimization (ECO-DQN) is, in principle, applicable to any combinatorial problem that can be defined on a graph. Experimentally, we show our method to produce state-of-the-art RL performance on the Maximum Cut problem. Moreover, because ECO-DQN can start from any arbitrary configuration, it can be combined with other search methods to further improve performance, which we demonstrate using a simple random search., Comment: In Proceedings of the 34th National Conference on Artificial Intelligence, AAAI 2020

Purcell-enhanced emission from a coupled emitter-cavity system is a fundamental manifestation of cavity quantum electrodynamics. Starting from a theoretical description we derive a scheme for photon emission from an emitter coupled to a birefringent cavity that exceeds hitherto anticipated limitations. Based on a recent study and experimental investigation of the intra-cavity coupling of orthogonal polarisation modes in birefringent cavities, we now decouple the emitter and the photon prior to emission from the cavity mode. Effectively, this is "hiding" the emitter from the photon in the cavity to suppress re-excitation, increasing the overall emission through the cavity mirrors. In doing so we show that tailored cavity birefringence can offer significant advantages and that these are practically achievable within the bounds of present-day technology. It is found that birefringence can mitigate the tradeoff between stronger emitter-cavity coupling and efficient photon extraction. This allows for longer cavities to be constructed without a loss of performance -- a significant result for applications where dielectric mirrors interfere with any trapping fields confining the emitter. We then generalise our model to consider a variety of equivalent schemes. For instance, detuning a pair of ground states in a three-level emitter coupled to a cavity in a Lambda-system is shown to provide the same enhancement, and it can be combined with a birefringent cavity to further increase performance. Additionally, it is found that when directly connecting multiple ground states of the emitter to form a chain of coupled states, the extraction efficiency approaches its fundamental upper limit. The principles proposed in this work can be applied in multiple ways to any emitter-cavity system, paving the way to surpassing the traditional limits of such systems with technologies that exist today., Comment: 8 pages, 8 figures plus 3 page appendix