Reaching out
Related Articles
Probing out-of-distribution generalization in machine learning for materials
Scientific machine learning (ML) aims to develop generalizable models, yet assessments of generalizability often rely on heuristics. Here, we demonstrate in the materials science setting that heuristic evaluations lead to biased conclusions of ML generalizability and benefits of neural scaling, through evaluations of out-of-distribution (OOD) tasks involving unseen chemistry or structural symmetries. Surprisingly, many tasks demonstrate good performance across models, including boosted trees. However, analysis of the materials representation space shows that most test data reside within regions well-covered by training data, while poorly-performing tasks involve data outside the training domain. For these challenging tasks, increasing training size or time yields limited or adverse effects, contrary to traditional neural scaling trends. Our findings highlight that most OOD tests reflect interpolation, not true extrapolation, leading to overestimations of generalizability and scaling benefits. This emphasizes the need for rigorously challenging OOD benchmarks.
Free mobility across group boundaries promotes intergroup cooperation
Group cooperation is a cornerstone of human society, enabling achievements that surpass individual capabilities. However, groups also define and restrict who benefits from cooperative actions and who does not, raising the question of how to foster cooperation across group boundaries. This study investigates the impact of voluntary mobility across group boundaries on intergroup cooperation. Participants, organized into two groups, decided whether to create benefits for themselves, group members, or everyone. In each round, they were paired with another participant and could reward the other’s actions during an ‘enforcement stage’, allowing for indirect reciprocity. In line with our preregistered hypothesis, when participants interacted only with in-group members, indirect reciprocity enforced group cooperation, while intergroup cooperation declined. Conversely, higher intergroup cooperation emerged when participants were forced to interact solely with out-group members. Crucially, in the free-mobility treatment – where participants could choose whether to meet an in-group or an out-group member in the enforcement stage – intergroup cooperation was significantly higher than when participants were forced to interact only with in-group members, even though most participants endogenously chose to interact with in-group members. A few ‘mobile individuals’ were sufficient to enforce intergroup cooperation by selectively choosing out-group members, enabling indirect reciprocity to transcend group boundaries. These findings highlight the importance of free intergroup mobility for overcoming the limitations of group cooperation.
Skill dependencies uncover nested human capital
Modern economies require increasingly diverse and specialized skills, many of which depend on the acquisition of other skills first. Here we analyse US survey data to reveal a nested structure within skill portfolios, where the direction of dependency is inferred from asymmetrical conditional probabilities—occupations require one skill conditional on another. This directional nature suggests that advanced, specific skills and knowledge are often built upon broader, fundamental ones. We examine 70 million job transitions to show that human capital development and career progression follow this structured pathway in which skills more aligned with the nested structure command higher wage premiums, require longer education and are less likely to be automated. These disparities are evident across genders and racial/ethnic groups, explaining long-term wage penalties. Finally, we find that this nested structure has become even more pronounced over the past two decades, indicating increased barriers to upward job mobility.
Annealing-inspired training of an optical neural network with ternary weights
Artificial neural networks (ANNs) represent a fundamentally connectionist and distributed approach to computing, and as such they differ from classical computers that utilize the von Neumann architecture. This has revived research interest in new unconventional hardware for more efficient ANNs rather than emulating them on traditional machines. To fully leverage ANNs, optimization algorithms must account for hardware limitations and imperfections. Photonics offers a promising platform with scalability, speed, energy efficiency, and parallel processing capabilities. However, fully autonomous optical neural networks (ONNs) with in-situ learning are scarce. In this work, we propose and demonstrate a ternary weight high-dimensional semiconductor laser-based ONN and introduce a method for achieving ternary weights using Boolean hardware, enhancing the ONN’s information processing capabilities. Furthermore, we design an in-situ optimization algorithm that is compatible with both Boolean and ternary weights. Our algorithm results in benefits, both in terms of convergence speed and performance. Our experimental results show the ONN’s long-term inference stability, with a consistency above 99% for over 10 h. Our work is of particular relevance in the context of in-situ learning under restricted hardware resources, especially since minimizing the power consumption of auxiliary hardware is crucial to preserving efficiency gains achieved by non-von Neumann ANN implementations.
Spin–valley protected Kramers pair in bilayer graphene
The intrinsic valley degree of freedom makes bilayer graphene (BLG) a unique platform for semiconductor qubits. The single-carrier quantum dot (QD) ground state exhibits a twofold degeneracy, where the two states that constitute a Kramers pair have opposite spin and valley quantum numbers. Because of the valley-dependent Berry curvature, an out-of-plane magnetic field breaks the time-reversal symmetry of this ground state and a qubit can be encoded in the spin–valley subspace. The Kramers states are protected against known spin- and valley-mixing mechanisms because mixing requires a simultaneous change of the two quantum numbers. Here, we fabricate a tunable QD device in Bernal BLG and measure a spin–valley relaxation time for the Kramers states of 38 s at 30 mK, which is two orders of magnitude longer than the 0.4 s measured for purely spin-blocked states. We also show that the intrinsic Kane–Mele spin–orbit splitting enables a Kramers doublet single-shot readout even at zero magnetic field with a fidelity above 99%. If these long-lived Kramers states also possess long coherence times and can be effectively manipulated, electrostatically defined QDs in BLG may serve as long-lived semiconductor qubits, extending beyond the spin qubit paradigm.
Responses