Position controllers have become the dominant interface for executing learned manipulation policies. Yet a critical design decision remains understudied: how should we choose controller gains for policy learning? The conventional wisdom is to select gains based on desired task compliance or stiffness. However, this logic breaks down when controllers are paired with state-conditioned policies: effective stiffness emerges from the interplay between learned reactions and control dynamics, not from gains alone. We argue that gain selection should instead be guided by learnability: how amenable different gain settings are to the learning algorithm in use. In this work, we systematically investigate how position controller gains affect three core components of modern robot learning pipelines: behavior cloning, reinforcement learning from scratch, and sim-to-real transfer. Through extensive experiments across multiple tasks and robot embodiments, we find that: (1) behavior cloning benefits from compliant and overdamped gain regimes, (2) reinforcement learning can succeed across all gain regimes given compatible hyperparameter tuning, and (3) sim-to-real transfer is harmed by stiff and ove
The task of predicting long-term patient outcomes using supervised machine learning is a challenging one, in part because of the high variance of each patient's trajectory, which can result in the model over-fitting to the training data. Temporal difference (TD) learning, a common reinforcement learning technique, may reduce variance by generalising learning to the pattern of state transitions rather than terminal outcomes. However, in healthcare this method requires several strong assumptions about patient states, and there appears to be limited literature evaluating the performance of TD learning against traditional supervised learning methods for long-term health outcome prediction tasks. In this study, we define a framework for applying TD learning to real-time irregularly sampled time series data using a Semi-Markov Reward Process. We evaluate the model framework in predicting intensive care mortality and show that TD learning under this framework can result in improved model robustness compared to standard supervised learning methods. and that this robustness is maintained even when validated on external datasets. This approach may offer a more reliable method when learning t
In an era of widespread web scraping, unlearnable dataset methods have the potential to protect data privacy by preventing deep neural networks from generalizing. But in addition to a number of practical limitations that make their use unlikely, we make a number of findings that call into question their ability to safeguard data. First, it is widely believed that neural networks trained on unlearnable datasets only learn shortcuts, simpler rules that are not useful for generalization. In contrast, we find that networks actually can learn useful features that can be reweighed for high test performance, suggesting that image protection is not assured. Unlearnable datasets are also believed to induce learning shortcuts through linear separability of added perturbations. We provide a counterexample, demonstrating that linear separability of perturbations is not a necessary condition. To emphasize why linearly separable perturbations should not be relied upon, we propose an orthogonal projection attack which allows learning from unlearnable datasets published in ICML 2021 and ICLR 2023. Our proposed attack is significantly less complex than recently proposed techniques.
Tool-augmented LLMs are increasingly deployed as agents that interleave natural-language reasoning with executable Python actions, as in CodeAct-style frameworks. In deployment, these agents rely on runtime state that persists across steps. By contrast, the traces used to post-train these models rarely encode how interpreter state is managed. We ask whether interpreter persistence is merely a runtime scaffold, or a property of the training data that shapes how agents learn to use the interpreter. We isolate state persistence as a training-time variable. We introduce Opaque Knapsack, a procedurally generated family of partially observable optimization tasks designed to prevent one-shot solutions. Item attributes and constraints are hidden behind budgeted tool calls, forcing multi-turn control flow and iterative state revision. Holding task instances, prompts, tools, model, and supervision fixed, we generate matched trajectories differing only in whether interpreter state persists across steps or resets after each action. We then fine-tune identical base models (Qwen3-8B) on each trace variant and evaluate all four train-runtime combinations. Our 2x2 cross-evaluation shows that inter
This study investigates how dynamical systems may be learned and modelled with a neuromorphic network which is itself a dynamical system. The neuromorphic network used in this study is based on a complex electrical circuit comprised of memristive elements that produce neuro-synaptic nonlinear responses to input electrical signals. To determine how computation may be performed using the physics of the underlying system, the neuromorphic network was simulated and evaluated on autonomous prediction of a multivariate chaotic time series, implemented with a reservoir computing framework. Through manipulating only input electrodes and voltages, optimal nonlinear dynamical responses were found when input voltages maximise the number of memristive components whose internal dynamics explore the entire dynamical range of the memristor model. Increasing the network coverage with the input electrodes was found to suppress other nonlinear responses that are less conducive to learning. These results provide valuable insights into how a physical neuromorphic network device can be feasibly optimised for learning complex dynamical systems using only external control parameters.
Scalable and generalizable physics-aware deep learning has long been considered a significant challenge with various applications across diverse domains ranging from robotics to molecular dynamics. Central to almost all physical systems are symplectic forms, the geometric backbone that underpins fundamental invariants like energy and momentum. In this work, we introduce a novel deep learning framework, MetaSym. In particular, MetaSym combines a strong symplectic inductive bias obtained from a symplectic encoder, and an autoregressive decoder with meta-attention. This principled design ensures that core physical invariants remain intact, while allowing flexible, data efficient adaptation to system heterogeneities. We benchmark MetaSym with highly varied and realistic datasets, such as a high-dimensional spring-mesh system Otness et al. (2021), an open quantum system with dissipation and measurement backaction, and robotics-inspired quadrotor dynamics. Crucially, we fine-tune and deploy MetaSym on real-world quadrotor data, demonstrating robustness to sensor noise and real-world uncertainty. Across all tasks, MetaSym achieves superior few-shot adaptation and outperforms larger state-
We present an ensemble machine-learning approach for composition-based, structure-agnostic screening of candidate superconductors among ternary hydrides under high pressure. Hydrogen-rich hydrides are known to exhibit high superconducting transition temperatures, and ternary or multinary hydrides can stabilize superconducting phases at reduced pressures through chemical compression. To systematically explore this vast compositional space, we construct an ensemble of 30 XGBoost regression models trained on a curated dataset of approximately 2000 binary and ternary hydride entries. The model ensemble is used to screen a broad set of A-B-H compositions at pressures of 100, 200, and 300 GPa, with screening outcomes evaluated statistically based on prediction consistency across ensemble members. This analysis highlights several high-scoring compositional systems, including Ca-Ti-H, Li-K-H, and Na-Mg-H, which were not explicitly included in the training dataset. In addition, feature-importance analysis indicates that elemental properties such as ionization energy and atomic radius contribute significantly to the learned composition-level trends in superconducting transition temperature.
We propose the use of the ``spin-opstring", derived from Stochastic Series Expansion Quantum Monte Carlo (QMC) simulations as machine learning (ML) input data. It offers a compact, memory-efficient representation of QMC simulation cells, combining the initial state with an operator string that encodes the state's evolution through imaginary time. Using supervised ML, we demonstrate the input's effectiveness in capturing both conventional and topological phase transitions, and in a regression task to predict non-local observables. We also demonstrate the capability of spin-opstring data in transfer learning by training models on one quantum system and successfully predicting on another, as well as showing that models trained on smaller system sizes generalize well to larger ones. Importantly, we illustrate a clear advantage of spin-opstring over conventional spin configurations in the accurate prediction of a quantum phase transition. Finally, we show how the inherent structure of spin-opstring provides an elegant framework for the interpretability of ML predictions. Using two state-of-the-art interpretability techniques, Layer-wise Relevance Propagation and SHapley Additive exPlana
Convolutional neural networks (CNNs) have become widely adopted in gravitational wave (GW) detection pipelines due to their ability to automatically learn hierarchical features from raw strain data. However, the physical meaning of these learned features remains underexplored, limiting the interpretability of such models. In this work, we propose a hybrid architecture that combines a CNN-based feature extractor with a random forest (RF) classifier to improve both detection performance and interpretability. Unlike prior approaches that directly connect classifiers to CNN outputs, our method introduces four physically interpretable metrics - variance, signal-to-noise ratio (SNR), waveform overlap, and peak amplitude - computed from the final convolutional layer. These are jointly used with the CNN output in the RF classifier to enable more informed decision boundaries. Tested on long-duration strain datasets, our hybrid model outperforms a baseline CNN model, achieving a relative improvement of 21\% in sensitivity at a fixed false alarm rate of 10 events per month. Notably, it also shows improved detection of low-SNR signals (SNR $\le$ 10), which are especially vulnerable to misclass
Offline reinforcement learning (RL) recovers the optimal policy $π$ given historical observations of an agent. In practice, $π$ is modeled as a weighted version of the agent's behavior policy $μ$, using a weight function $w$ working as a critic of the agent's behavior. Though recent approaches to offline RL based on diffusion models have exhibited promising results, the computation of the required scores is challenging due to their dependence on the unknown $w$. In this work, we alleviate this issue by constructing a diffusion over both the actions and the weights. With the proposed setting, the required scores are directly obtained from the diffusion model without learning extra networks. Our main conceptual contribution is a novel guidance method, where guidance (which is a function of $w$) comes from the same diffusion model, therefore, our proposal is termed Self-Weighted Guidance (SWG). We show that SWG generates samples from the desired distribution on toy examples and performs on par with state-of-the-art methods on D4RL's challenging environments, while maintaining a streamlined training pipeline. We further validate SWG through ablation studies on weight formulations and s
Evaluation of machine learning models typically emphasizes final accuracy, overlooking the cost of adaptation: the cumulative errors incurred while learning from scratch. Guess-and- Learn (G&L) v1.0 addresses this gap by measuring cold-start adaptability - the total mistakes a model makes while sequentially labeling an unlabeled dataset. At each step, the learner selects an instance, predicts its label, receives the ground truth, and updates parameters under either online (per-sample) or batch (delayed) mode. The resulting error trajectory exposes adaptation speed, selection quality, and bias - dynamics invisible to endpoint metrics. G&L defines four tracks (Scratch/Pretrained $\times$ Online/Batch) to disentangle the effects of initialization and update frequency. We formalize the protocol, relate it to classical mistake-bound theory, and estimate a heuristic "oracle reference band" for MNIST as a plausibility reference. Baseline experiments on MNIST and AG News, spanning classical methods (Perceptron, k-NN), convolutional architectures (CNN, ResNet-50), and pretrained transformers (ViT-B/16, BERT-base), reveal systematic differences in early-phase efficiency: smaller mode
Deep learning-based approaches for generating novel drug molecules with specific properties have gained a lot of interest in the last few years. Recent studies have demonstrated promising performance for string-based generation of novel molecules utilizing reinforcement learning. In this paper, we develop a unified framework for using reinforcement learning for de novo drug design, wherein we systematically study various on- and off-policy reinforcement learning algorithms and replay buffers to learn an RNN-based policy to generate novel molecules predicted to be active against the dopamine receptor DRD2. Our findings suggest that it is advantageous to use at least both top-scoring and low-scoring molecules for updating the policy when structural diversity is essential. Using all generated molecules at an iteration seems to enhance performance stability for on-policy algorithms. In addition, when replaying high, intermediate, and low-scoring molecules, off-policy algorithms display the potential of improving the structural diversity and number of active molecules generated, but possibly at the cost of a longer exploration phase. Our work provides an open-source framework enabling r
In recent years, machine learning models, chiefly deep neural networks, have revealed suited to learn accurate energy-density functionals from data. However, problematic instabilities have been shown to occur in the search of ground-state density profiles via energy minimization. Indeed, any small noise can lead astray from realistic profiles, causing the failure of the learned functional and, hence, strong violations of the variational property. In this article, we employ variational autoencoders to build a compressed, flexible, and regular representation of the ground-state density profiles of various quantum models. Performing energy minimization in this compressed space allows us to avoid both numerical instabilities and variational biases due to excessive constraints. Our tests are performed on one-dimensional single-particle models from the literature in the field and, notably, on a three-dimensional disordered potential. In all cases, the ground-state energies are estimated with errors below the chemical accuracy and the density profiles are accurately reproduced without numerical artifacts.
Entanglement is fundamental to quantum information science and technology, yet controlling and manipulating entanglement -- so-called entanglement engineering -- for arbitrary quantum systems remains a formidable challenge. There are two difficulties: the fragility of quantum entanglement and its experimental characterization. We develop a model-free deep reinforcement-learning (RL) approach to entanglement engineering, in which feedback control together with weak continuous measurement and partial state observation is exploited to generate and maintain desired entanglement. We employ quantum optomechanical systems with linear or nonlinear photon-phonon interactions to demonstrate the workings of our machine-learning-based entanglement engineering protocol. In particular, the RL agent sequentially interacts with one or multiple parallel quantum optomechanical environments, collects trajectories, and updates the policy to maximize the accumulated reward to create and stabilize quantum entanglement over an arbitrary amount of time. The machine-learning-based model-free control principle is applicable to the entanglement engineering of experimental quantum systems in general.
Effectively supporting students in mastering all facets of self-regulated learning is a central aim of teachers and educational researchers. Prior research could demonstrate that formative feedback is an effective way to support students during self-regulated learning (SRL). However, for formative feedback to be effective, it needs to be tailored to the learners, requiring information about their learning progress. In this work, we introduce LEAP, a novel platform that utilizes advanced large language models (LLMs), such as ChatGPT, to provide formative feedback to students. LEAP empowers teachers with the ability to effectively pre-prompt and assign tasks to the LLM, thereby stimulating students' cognitive and metacognitive processes and promoting self-regulated learning. We demonstrate that a systematic prompt design based on theoretical principles can provide a wide range of types of scaffolds to students, including sense-making, elaboration, self-explanation, partial task-solution scaffolds, as well as metacognitive and motivational scaffolds. In this way, we emphasize the critical importance of synchronizing educational technological advances with empirical research and theore
Discovery of the molecular candidates for applications in drug targets, biomolecular systems, catalysts, photovoltaics, organic electronics, and batteries, necessitates development of machine learning algorithms capable of rapid exploration of the chemical spaces targeting the desired functionalities. Here we introduce a novel approach for the active learning over the chemical spaces based on hypothesis learning. We construct the hypotheses on the possible relationships between structures and functionalities of interest based on a small subset of data and introduce them as (probabilistic) mean functions for the Gaussian process. This approach combines the elements from the symbolic regression methods such as SISSO and active learning into a single framework. The primary focus of constructing this framework is to approximate physical laws in an active learning regime toward a more robust predictive performance, as traditional evaluation on hold-out sets in machine learning doesn't account for out-of-distribution effects and may lead to a complete failure on unseen chemical space. Here, we demonstrate it for the QM9 dataset, but it can be applied more broadly to datasets from both do
Conventional rule learning algorithms aim at finding a set of simple rules, where each rule covers as many examples as possible. In this paper, we argue that the rules found in this way may not be the optimal explanations for each of the examples they cover. Instead, we propose an efficient algorithm that aims at finding the best rule covering each training example in a greedy optimization consisting of one specialization and one generalization loop. These locally optimal rules are collected and then filtered for a final rule set, which is much larger than the sets learned by conventional rule learning algorithms. A new example is classified by selecting the best among the rules that cover this example. In our experiments on small to very large datasets, the approach's average classification accuracy is higher than that of state-of-the-art rule learning algorithms. Moreover, the algorithm is highly efficient and can inherently be processed in parallel without affecting the learned rule set and so the classification accuracy. We thus believe that it closes an important gap for large-scale classification rule induction.
We introduce SynCLR, a novel approach for learning visual representations exclusively from synthetic images and synthetic captions, without any real data. We synthesize a large dataset of image captions using LLMs, then use an off-the-shelf text-to-image model to generate multiple images corresponding to each synthetic caption. We perform visual representation learning on these synthetic images via contrastive learning, treating images sharing the same caption as positive pairs. The resulting representations transfer well to many downstream tasks, competing favorably with other general-purpose visual representation learners such as CLIP and DINO v2 in image classification tasks. Furthermore, in dense prediction tasks such as semantic segmentation, SynCLR outperforms previous self-supervised methods by a significant margin, e.g., improving over MAE and iBOT by 6.2 and 4.3 mIoU on ADE20k for ViT-B/16.
This paper presents a learning framework to estimate an agent capability and task requirement model for multi-agent task allocation. With a set of team configurations and the corresponding task performances as the training data, linear task constraints can be learned to be embedded in many existing optimization-based task allocation frameworks. Comprehensive computational evaluations are conducted to test the scalability and prediction accuracy of the learning framework with a limited number of team configurations and performance pairs. A ROS and Gazebo-based simulation environment is developed to validate the proposed requirements learning and task allocation framework in practical multi-agent exploration and manipulation tasks. Results show that the learning process for scenarios with 40 tasks and 6 types of agents uses around 12 seconds, ending up with prediction errors in the range of 0.5-2%.
The inherent trade-off in on-line learning is between exploration and exploitation. A good balance between these two (conflicting) goals can achieve a better long-term performance. Can we define an optimal balance? We propose to study this question through a backward-adaptive lossy compression system, which exhibits a "natural" trade-off between exploration and exploitation.