Search

We introduce the concept of the self-referencing causal cycle (abbreviated RECALL) - a mechanism that enables large language models (LLMs) to bypass the limitations of unidirectional causality, which underlies a phenomenon known as the reversal curse. When an LLM is prompted with sequential data, it often fails to recall preceding context. For example, when we ask an LLM to recall the line preceding "O say does that star-spangled banner yet wave" in the U.S. National Anthem, it often fails to correctly return "Gave proof through the night that our flag was still there" - this is due to the reversal curse. It occurs because language models such as ChatGPT and Llama generate text based on preceding tokens, requiring facts to be learned and reproduced in a consistent token order. While the reversal curse is often viewed as a limitation, we offer evidence of an alternative view: it is not always an obstacle in practice. We find that RECALL is driven by what we designate as cycle tokens - sequences that connect different parts of the training data, enabling recall of preceding tokens from succeeding ones. Through rigorous probabilistic formalization and controlled experiments, we demonstrate how the cycles they induce influence a model's ability to reproduce information. To facilitate reproducibility, we provide our code and experimental details at https://anonymous.4open.science/r/remember-B0B8/.

Ferrimagnetic materials with a compensation temperature have recently attracted interest because of their unique combination of ferromagnetic and antiferromagnetic properties. However, their magnetization dynamics near the compensation temperature are complex and cannot be fully explained by conventional ferromagnetic resonance (FMR) or exchange resonance modes. Therefore, practical models are necessary to capture these dynamics accurately. In this study, we derived the analytical solutions for the magnetic resonance frequencies of compensated ferrimagnets over all temperature ranges, considering both the in-plane and out-of-plane orientations of the magnetization. Our solutions successfully reproduce the experimental data obtained from time-resolved magneto-optical Faraday rotation and Brillouin light scattering measurements for the in-plane and out-of-plane cases, respectively. This reproduction is achieved by incorporating the exchange stiffness and temperature dependence of the magnetic anisotropy into the free energy density. Additionally, at temperatures sufficiently far from the compensation temperature, our analytical solutions converge with the conventional FMR and exchange resonance models.

Although 3D object editing has the potential to significantly influence various industries, recent research in 3D generation and editing has primarily focused on converting text and images into 3D models, often overlooking the need for fine-grained control over the editing of existing 3D objects. This paper introduces a framework that employs a pre-trained regressor, enabling continuous, precise, attribute-specific modifications to both the stylistic and geometric attributes of vehicle 3D models. Our method not only preserves the inherent identity of vehicle 3D objects, but also supports multi-attribute editing, allowing for extensive customization without compromising the model's structural integrity. Experimental results demonstrate the efficacy of our approach in achieving detailed edits on various vehicle 3D models., Comment: 18 pages, 12 figures

The XLZD collaboration is developing a two-phase xenon time projection chamber with an active mass of 60 to 80 t capable of probing the remaining WIMP-nucleon interaction parameter space down to the so-called neutrino fog. In this work we show that, based on the performance of currently operating detectors using the same technology and a realistic reduction of radioactivity in detector materials, such an experiment will also be able to competitively search for neutrinoless double beta decay in $^{136}$Xe using a natural-abundance xenon target. XLZD can reach a 3$\sigma$ discovery potential half-life of 5.7$\times$10$^{27}$ yr (and a 90% CL exclusion of 1.3$\times$10$^{28}$ yr) with 10 years of data taking, corresponding to a Majorana mass range of 7.3-31.3 meV (4.8-20.5 meV). XLZD will thus exclude the inverted neutrino mass ordering parameter space and will start to probe the normal ordering region for most of the nuclear matrix elements commonly considered by the community., Comment: 29 pages, 7 figures

This report describes the experimental strategy and technologies for a next-generation xenon observatory sensitive to dark matter and neutrino physics. The detector will have an active liquid xenon target mass of 60-80 tonnes and is proposed by the XENON-LUX-ZEPLIN-DARWIN (XLZD) collaboration. The design is based on the mature liquid xenon time projection chamber technology of the current-generation experiments, LZ and XENONnT. A baseline design and opportunities for further optimization of the individual detector components are discussed. The experiment envisaged here has the capability to explore parameter space for Weakly Interacting Massive Particle (WIMP) dark matter down to the neutrino fog, with a 3$\sigma$ evidence potential for the spin-independent WIMP-nucleon cross sections as low as $3\times10^{-49}\rm cm^2$ (at 40 GeV/c$^2$ WIMP mass). The observatory is also projected to have a 3$\sigma$ observation potential of neutrinoless double-beta decay of $^{136}$Xe at a half-life of up to $5.7\times 10^{27}$ years. Additionally, it is sensitive to astrophysical neutrinos from the atmosphere, sun, and galactic supernovae., Comment: 32 pages, 14 figures

This paper introduces repetition neurons, regarded as skill neurons responsible for the repetition problem in text generation tasks. These neurons are progressively activated more strongly as repetition continues, indicating that they perceive repetition as a task to copy the previous context repeatedly, similar to in-context learning. We identify these repetition neurons by comparing activation values before and after the onset of repetition in texts generated by recent pre-trained language models. We analyze the repetition neurons in three English and one Japanese pre-trained language models and observe similar patterns across them.

This paper investigates whether large language models (LLMs) utilize numerical attributes encoded in a low-dimensional subspace of the embedding space when answering logical comparison questions (e.g., Was Cristiano born before Messi?). We first identified these subspaces using partial least squares regression, which effectively encodes the numerical attributes associated with the entities in comparison prompts. Further, we demonstrate causality by intervening in these subspaces to manipulate hidden states, thereby altering the LLM's comparison outcomes. Experimental results show that our findings hold for different numerical attributes, indicating that LLMs utilize the linearly encoded information for numerical reasoning.

This paper proposes two quantum operation scheduling methods for accelerating parallel state-vector-based quantum circuit simulation using multiple graphics processing units (GPUs). The proposed methods reduce all-to-all communication caused by qubit reordering (QR), which can dominate the overhead of parallel simulation. Our approach eliminates redundant QRs by introducing intentional delays in QR communications such that multiple QRs can be aggregated into a single QR. The delays are carefully introduced based on the principles of time-space tiling, or a cache optimization technique for classical computers, which we use to arrange the execution order of quantum operations. Moreover, we present an extended scheduling method for the hierarchical interconnection of GPU cluster systems to avoid slow inter-node communication. We develop these methods tailored for two primary procedures in variational quantum eigensolver (VQE) simulation: quantum state update (QSU) and expectation value computation (EVC). Experimental validation on 32-GPU executions demonstrates acceleration in QSU and EVC -- up to 54$\times$ and 606$\times$, respectively -- compared to existing methods. Moreover, our extended scheduling method further reduced communication time by up to 15\% in a two-layered interconnected cluster system. Our approach is useful for any quantum circuit simulations, including QSU and/or EVC., Comment: 24 pages, 18 figures

Place names, or toponyms, play an integral role in human representation and communication of geographic space. In particular, how people relate each toponym with particular locations in geographic space should be indicative of their spatial perception. Here, we make use of an extensive dataset of georeferenced social media posts, retreived from Twitter, to perform a statistical analysis of the geographic distribution of toponyms and uncover the relationship between toponyms and geographic space. We show that the occurrence of toponyms is characterized by spatial inhomogeneity, giving rise to patterns that are distinct from the distribution of common nouns. Using simple models, we quantify the spatial specificity of toponym distributions and identify their core-periphery structures. In particular, we find that toponyms are used with a probability that decays as a power law with distance from the geographic center of their occurrence. Our findings highlight the potential of social media data to explore linguistic patterns in geographic space, paving the way for comprehensive analyses of human spatial representations., Comment: Main text: 13 pages, 7 figures. Supporting information: 5 pages, 5 figures

We present a novel deep learning pipeline to perform a model-independent, likelihood-free search for anomalous (i.e., non-background) events in the proposed next generation multi-ton scale liquid Xenon-based direct detection experiment, DARWIN. We train an anomaly detector comprising a variational autoencoder and a classifier on extensive, high-dimensional simulated detector response data and construct a one-dimensional anomaly score optimised to reject the background only hypothesis in the presence of an excess of non-background-like events. We benchmark the procedure with a sensitivity study that determines its power to reject the background-only hypothesis in the presence of an injected WIMP dark matter signal, outperforming the classical, likelihood-based background rejection test. We show that our neural networks learn relevant energy features of the events from low-level, high-dimensional detector outputs, without the need to compress this data into lower-dimensional observables, thus reducing computational effort and information loss. For the future, our approach lays the foundation for an efficient end-to-end pipeline that eliminates the need for many of the corrections and cuts that are traditionally part of the analysis chain, with the potential of achieving higher accuracy and significant reduction of analysis time., Comment: 10 Figures, 3 Tables, 23 Pages (incl. references)

A multiplication on persistence diagrams is introduced by means of Schubert calculus. The key observation behind this multiplication comes from the fact that the representation space of persistence modules has the structure of the Schubert decomposition of a flag. In particular, isomorphism classes of persistence modules correspond to Schubert cells, thereby the Schubert calculus naturally defines a multiplication on persistence diagrams. The meaning of the multiplication on persistence diagrams is carried over from that on Schubert calculus, i.e., algebro-geometric intersections of varieties of persistence modules.

Band gap energy of an organic molecule can be reduced by intermolecular interaction enhancement, and thus, certain polycyclic aromatic hydrocarbons (PAHs), which are insulators with wide band gaps, are expected to undergo insulator-metal transitions by simple compression. Such a pressure-induced electronic transition can be exploited to transform non-metallic organic materials into states featuring intriguing electronic characteristics such as high-temperature superconductivity. Numerous attempts have been made to metalize various small PAHs, but so far only pressure-induced amorphization well below the megabar region was observed. The wide band gap energy of the small PAHs and low chemical stability under simple compression are the bottlenecks. We have investigated the band gap energy evolution and the crystal structural compression of the large PAH molecules, where the band gap energy is significantly reduced by increasing the number of {\pi}-electrons and improved chemical stability with fully benzenoid molecular structure. Herein, we present a pressure-induced transition in dicoronylene, C48H20, an insulator at ambient conditions that transforms into a semi-metallic state above 23.0 GPa with a three-order-of-magnitude reduction in resistivity. In-situ UV-visible absorption, transport property measurement, Raman spectroscopy, X-ray diffraction and density functional theory calculations were performed to provide tentative explanations to the alterations in its electronic structure at high pressure. The discovery of an electronic transition at pressures well below the megabar is a promising step towards realization of a single component purely hydrocarbon molecular metal in the near future., Comment: 22 pages, 5 figures

Many datasets of natural language processing (NLP) sometimes include annotation errors. Researchers have attempted to develop methods to reduce the adverse effect of errors in datasets automatically. However, an existing method is time-consuming because it requires many trained models to detect errors. We propose a novel method to reduce the time of error detection. Specifically, we use a tokenization technique called subword regularization to create pseudo-multiple models which are used to detect errors. Our proposed method, SubRegWeigh, can perform annotation weighting four to five times faster than the existing method. Additionally, SubRegWeigh improved performance in both document classification and named entity recognition tasks. In experiments with pseudo-incorrect labels, pseudo-incorrect labels were adequately detected., Comment: 14 pages, 1 figures, 10 tables

The estimation and management of motor temperature are important for the continuous movements of robots. In this study, we propose an online learning method of thermal model parameters of motors for an accurate estimation of motor core temperature. Also, we propose a management method of motor core temperature using the updated model and anomaly detection method of motors. Finally, we apply this method to the muscles of the musculoskeletal humanoid and verify the ability of continuous movements., Comment: Accepted at IEEE Robotics and Automation Letters

This paper introduces LLM-jp, a cross-organizational project for the research and development of Japanese large language models (LLMs). LLM-jp aims to develop open-source and strong Japanese LLMs, and as of this writing, more than 1,500 participants from academia and industry are working together for this purpose. This paper presents the background of the establishment of LLM-jp, summaries of its activities, and technical reports on the LLMs developed by LLM-jp. For the latest activities, visit https://llm-jp.nii.ac.jp/en/.

In reinforcement learning (RL) with experience replay, experiences stored in a replay buffer influence the RL agent's performance. Information about how these experiences influence the agent's performance is valuable for various purposes, such as identifying experiences that negatively influence underperforming agents. One method for estimating the influence of experiences is the leave-one-out (LOO) method. However, this method is usually computationally prohibitive. In this paper, we present Policy Iteration with Turn-over Dropout (PIToD), which efficiently estimates the influence of experiences. We evaluate how accurately PIToD estimates the influence of experiences and its efficiency compared to LOO. We then apply PIToD to amend underperforming RL agents, i.e., we use PIToD to estimate negatively influential experiences for the RL agents and to delete the influence of these experiences. We show that RL agents' performance is significantly improved via amendments with PIToD., Comment: Source code: https://github.com/TakuyaHiraoka/Which-Experiences-Are-Influential-for-RL-Agents

Library-based methods are known to be very effective for fast motion planning by adapting an experience retrieved from a precomputed library. This article presents CoverLib, a principled approach for constructing and utilizing such a library. CoverLib iteratively adds an experience-classifier-pair to the library, where each classifier corresponds to an adaptable region of the experience within the problem space. This iterative process is an active procedure, as it selects the next experience based on its ability to effectively cover the uncovered region. During the query phase, these classifiers are utilized to select an experience that is expected to be adaptable for a given problem. Experimental results demonstrate that CoverLib effectively mitigates the trade-off between plannability and speed observed in global (e.g. sampling-based) and local (e.g. optimization-based) methods. As a result, it achieves both fast planning and high success rates over the problem domain. Moreover, due to its adaptation-algorithm-agnostic nature, CoverLib seamlessly integrates with various adaptation methods, including nonlinear programming-based and sampling-based algorithms.

The stability of persistent homology has led to wide applications of the persistence diagram as a trusted topological descriptor in the presence of noise. However, with the increasing demand for high-dimension and low-sample-size data processing in modern science, it is questionable whether persistence diagrams retain their reliability in the presence of high-dimensional noise. This work aims to study the reliability of persistence diagrams in the high-dimension low-sample-size data setting. By analyzing the asymptotic behavior of persistence diagrams for high-dimensional random data, we show that persistence diagrams are no longer reliable descriptors of low-sample-size data under high-dimensional noise perturbations. We refer to this loss of reliability of persistence diagrams in such data settings as the curse of dimensionality on persistence diagrams. Next, we investigate the possibility of using normalized principal component analysis as a method for reducing the dimensionality of the high-dimensional observed data to resolve the curse of dimensionality. We show that this method can mitigate the curse of dimensionality on persistence diagrams. Our results shed some new light on the challenges of processing high-dimension low-sample-size data by persistence diagrams and provide a starting point for future research in this area.

We explore threshold vocabulary trimming in Byte-Pair Encoding subword tokenization, a postprocessing step that replaces rare subwords with their component subwords. The technique is available in popular tokenization libraries but has not been subjected to rigorous scientific scrutiny. While the removal of rare subwords is suggested as best practice in machine translation implementations, both as a means to reduce model size and for improving model performance through robustness, our experiments indicate that, across a large space of hyperparameter settings, vocabulary trimming fails to improve performance, and is even prone to incurring heavy degradation., Comment: 15 pages

Large language models (LLMs) have increased interest in vision language models (VLMs), which process image-text pairs as input. Studies investigating the visual understanding ability of VLMs have been proposed, but such studies are still preliminary because existing datasets do not permit a comprehensive evaluation of the fine-grained visual linguistic abilities of VLMs across multiple languages. To further explore the strengths of VLMs, such as GPT-4V \cite{openai2023GPT4}, we developed new datasets for the systematic and qualitative analysis of VLMs. Our contribution is four-fold: 1) we introduced nine vision-and-language (VL) tasks (including object recognition, image-text matching, and more) and constructed multilingual visual-text datasets in four languages: English, Japanese, Swahili, and Urdu through utilizing templates containing \textit{questions} and prompting GPT4-V to generate the \textit{answers} and the \textit{rationales}, 2) introduced a new VL task named \textit{unrelatedness}, 3) introduced rationales to enable human understanding of the VLM reasoning process, and 4) employed human evaluation to measure the suitability of proposed datasets for VL tasks. We show that VLMs can be fine-tuned on our datasets. Our work is the first to conduct such analyses in Swahili and Urdu. Also, it introduces \textit{rationales} in VL analysis, which played a vital role in the evaluation.

Do pretrained language models have knowledge regarding the surface information of tokens? We examined the surface information stored in word or subword embeddings acquired by pretrained language models from the perspectives of token length, substrings, and token constitution. Additionally, we evaluated the ability of models to generate knowledge regarding token surfaces. We focused on 12 pretrained language models that were mainly trained on English and Japanese corpora. Experimental results demonstrate that pretrained language models have knowledge regarding token length and substrings but not token constitution. Additionally, the results imply that there is a bottleneck on the decoder side in terms of effectively utilizing acquired knowledge.

Transferring human motion skills to humanoid robots remains a significant challenge. In this study, we introduce a Wasserstein adversarial imitation learning system, allowing humanoid robots to replicate natural whole-body locomotion patterns and execute seamless transitions by mimicking human motions. First, we present a unified primitive-skeleton motion retargeting to mitigate morphological differences between arbitrary human demonstrators and humanoid robots. An adversarial critic component is integrated with Reinforcement Learning (RL) to guide the control policy to produce behaviors aligned with the data distribution of mixed reference motions. Additionally, we employ a specific Integral Probabilistic Metric (IPM), namely the Wasserstein-1 distance with a novel soft boundary constraint to stabilize the training process and prevent mode collapse. Our system is evaluated on a full-sized humanoid JAXON in the simulator. The resulting control policy demonstrates a wide range of locomotion patterns, including standing, push-recovery, squat walking, human-like straight-leg walking, and dynamic running. Notably, even in the absence of transition motions in the demonstration dataset, robots showcase an emerging ability to transit naturally between distinct locomotion patterns as desired speed changes.

We performed time-resolved pump--probe measurements using rare-earth iron garnet \ce{Gd3/2Yb1/2BiFe5O12} as a two-sublattice ferrimagnet. We measured the initial phases of the magnetic resonance modes below and above the magnetization compensation temperature to clarify the sublattice selectivity of the inverse Faraday effect in ferrimagnets. A comparison of the time evolution of magnetization estimated using the equations of motion revealed that the inverse Faraday effect occurring in ferrimagnetic materials has sublattice selectivity. This is in striking contrast to antiferromagnets, in which the inverse Faraday effect acts on each sublattice identically. The initial phase analysis can be applied to other ferrimagnets with compensation temperatures., Comment: 4 pages, 5 figures

Reinforcement learning (RL) methods with a high replay ratio (RR) and regularization have gained interest due to their superior sample efficiency. However, these methods have mainly been developed for dense-reward tasks. In this paper, we aim to extend these RL methods to sparse-reward goal-conditioned tasks. We use Randomized Ensemble Double Q-learning (REDQ) (Chen et al., 2021), an RL method with a high RR and regularization. To apply REDQ to sparse-reward goal-conditioned tasks, we make the following modifications to it: (i) using hindsight experience replay and (ii) bounding target Q-values. We evaluate REDQ with these modifications on 12 sparse-reward goal-conditioned tasks of Robotics (Plappert et al., 2018), and show that it achieves about $2 \times$ better sample efficiency than previous state-of-the-art (SoTA) RL methods. Furthermore, we reconsider the necessity of specific components of REDQ and simplify it by removing unnecessary ones. The simplified REDQ with our modifications achieves $\sim 8 \times$ better sample efficiency than the SoTA methods in 4 Fetch tasks of Robotics., Comment: Source code: https://github.com/TakuyaHiraoka/Efficient-SRGC-RL-with-a-High-RR-and-Regularization Demo video: https://drive.google.com/file/d/1UHd7JVPCwFLNFhy1QcycQfwU_nll_yII/view?usp=drive_link