Search

The Simons Observatory (SO) is a ground-based cosmic microwave background (CMB) survey experiment that currently consists of three 0.42m small-aperture telescopes (SATs) and one 6m large-aperture telescope (LAT), located at an elevation of 5200m in the Atacama Desert in Chile. At the LAT's focal plane, SO will install >62,000 transition-edge sensor detectors across 13 optics tubes (OTs) within the Large Aperture Telescope Receiver (LATR), the largest cryogenic camera ever built to observe the CMB. Here we report on the validation of the LATR in the laboratory and the subsequent dark testing and validation within the LAT. We show that the LATR meets cryogenic, optical, and detector specifications required for high-sensitivity measurements of the CMB. At the time of writing, the LATR is installed in the LAT with six OTs (corresponding to >31,000 detectors), and the LAT mirrors and remaining seven OTs are undergoing development.

This report arrives during a time of heightened scrutiny and civic engagement. While public attention is currently trained primarily on Milwaukee Public Schools (MPS), this research encompasses the whole of the city's education system, marking both commonalities and differences between different types of publicly funded schools. It highlights financial aspects of the system but also expands the discussion to include student enrollment and outcomes. The report is organized into three major sections: Milwaukee's Students and Schools, covering the basic schooling options available to students along with student enrollment and demographic trends; School Funding in Milwaukee, outlining the core funding mechanisms and their funding levels for schools in the city; and Student Outcomes and School Performance, summarizing broad trends in Milwaukee's academic results. The report concludes with key insights for the consideration of both policymakers and the public. It is hoped the report's findings ground both current and future policy discussions with important facts and nonpartisan insights. The authors further intend to inform those discussions with a second report this fall, which will highlight promising K-12 practices and innovations occurring both within Milwaukee and nationally.

Model-based reinforcement learning algorithms that combine model-based planning and learned value/policy prior have gained significant recognition for their high data efficiency and superior performance in continuous control. However, we discover that existing methods that rely on standard SAC-style policy iteration for value learning, directly using data generated by the planner, often result in \emph{persistent value overestimation}. Through theoretical analysis and experiments, we argue that this issue is deeply rooted in the structural policy mismatch between the data generation policy that is always bootstrapped by the planner and the learned policy prior. To mitigate such a mismatch in a minimalist way, we propose a policy regularization term reducing out-of-distribution (OOD) queries, thereby improving value learning. Our method involves minimum changes on top of existing frameworks and requires no additional computation. Extensive experiments demonstrate that the proposed approach improves performance over baselines such as TD-MPC2 by large margins, particularly in 61-DoF humanoid tasks. View qualitative results at https://darthutopian.github.io/tdmpc_square/.

Recent technological advances have introduced new high-throughput methods for studying host-virus interactions, but testing synergistic interactions between host gene pairs during infection remains relatively slow and labor intensive. Identification of multiple gene knockdowns that effectively inhibit viral replication requires a search over the combinatorial space of all possible target gene pairs and is infeasible via brute-force experiments. Although active learning methods for sequential experimental design have shown promise, existing approaches have generally been restricted to single-gene knockdowns or small-scale double knockdown datasets. In this study, we present an integrated Deep Active Learning (DeepAL) framework that incorporates information from a biological knowledge graph (SPOKE, the Scalable Precision Medicine Open Knowledge Engine) to efficiently search the configuration space of a large dataset of all pairwise knockdowns of 356 human genes in HIV infection. Through graph representation learning, the framework is able to generate task-specific representations of genes while also balancing the exploration-exploitation trade-off to pinpoint highly effective double-knockdown pairs. We additionally present an ensemble method for uncertainty quantification and an interpretation of the gene pairs selected by our algorithm via pathway analysis. To our knowledge, this is the first work to show promising results on double-gene knockdown experimental data of appreciable scale (356 by 356 matrix).

Artistic creation is often seen as a uniquely human endeavor, yet robots bring distinct advantages to music-making, such as precise tempo control, unpredictable rhythmic complexities, and the ability to coordinate intricate human and robot performances. While many robotic music systems aim to mimic human musicianship, our work emphasizes the unique strengths of robots, resulting in a novel multi-robot performance instrument called the Beatbots, capable of producing music that is challenging for humans to replicate using current methods. The Beatbots were designed using an ``informed prototyping'' process, incorporating feedback from three musicians throughout development. We evaluated the Beatbots through a live public performance, surveying participants (N=28) to understand how they perceived and interacted with the robotic performance. Results show that participants valued the playfulness of the experience, the aesthetics of the robot system, and the unconventional robot-generated music. Expert musicians and non-expert roboticists demonstrated especially positive mindset shifts during the performance, although participants across all demographics had favorable responses. We propose design principles to guide the development of future robotic music systems and identify key robotic music affordances that our musician consultants considered particularly important for robotic music performance., Comment: Copyright protected by IEEE/ACM, 10 pages, 4 figures, 1 table, in proceedings of 20th IEEE/ACM International Conference on Human-Robot Interaction (HRI 2025)

Astronomical radio interferometers achieve exquisite angular resolution by cross-correlating signal from a cosmic source simultaneously observed by distant pairs of radio telescopes to produce a Fourier-type measurement called a visibility. Million Points of Light (MPoL) is a Python library supporting feed-forward modeling of interferometric visibility datasets for synthesis imaging and parametric Bayesian inference, built using the autodifferentiable machine learning framework PyTorch. Neural network components provide a rich set of modular and composable building blocks that can be used to express the physical relationships between latent model parameters and observed data following the radio interferometric measurement equation. Industry-grade optimizers make it straightforward to simultaneously solve for the synthesized image and calibration parameters using stochastic gradient descent., Comment: Submitted to JOSS. 1559 words, 1 figure

We present and release MIDI-GPT, a generative system based on the Transformer architecture that is designed for computer-assisted music composition workflows. MIDI-GPT supports the infilling of musical material at the track and bar level, and can condition generation on attributes including: instrument type, musical style, note density, polyphony level, and note duration. In order to integrate these features, we employ an alternative representation for musical material, creating a time-ordered sequence of musical events for each track and concatenating several tracks into a single sequence, rather than using a single time-ordered sequence where the musical events corresponding to different tracks are interleaved. We also propose a variation of our representation allowing for expressiveness. We present experimental results that demonstrate that MIDI-GPT is able to consistently avoid duplicating the musical material it was trained on, generate music that is stylistically similar to the training dataset, and that attribute controls allow enforcing various constraints on the generated material. We also outline several real-world applications of MIDI-GPT, including collaborations with industry partners that explore the integration and evaluation of MIDI-GPT into commercial products, as well as several artistic works produced using it., Comment: AAAI 25

Mechanistic interpretability aims to understand the computational mechanisms underlying neural networks' capabilities in order to accomplish concrete scientific and engineering goals. Progress in this field thus promises to provide greater assurance over AI system behavior and shed light on exciting scientific questions about the nature of intelligence. Despite recent progress toward these goals, there are many open problems in the field that require solutions before many scientific and practical benefits can be realized: Our methods require both conceptual and practical improvements to reveal deeper insights; we must figure out how best to apply our methods in pursuit of specific goals; and the field must grapple with socio-technical challenges that influence and are influenced by our work. This forward-facing review discusses the current frontier of mechanistic interpretability and the open problems that the field may benefit from prioritizing.

Large language models (LLMs) have demonstrated immense potential across various tasks. However, research for exploring and improving the capabilities of LLMs in interpreting graph structures remains limited. To address this gap, we conduct a comprehensive evaluation of prompting current open-source LLMs on graph-to-text generation tasks. Although we explored the optimal prompting strategies and proposed a novel and effective diversity-difficulty-based few-shot sample selection method, we found that the improvements from tuning-free approaches were incremental, as LLMs struggle with planning on complex graphs, particularly those with a larger number of triplets. To further improve LLMs in planning with graph sequences and grounding in truth, we introduce a new graph-to-text dataset, PlanGTG, annotated with two sub-tasks: reordering and attribution. Through extensive automatic and human evaluations, we demonstrate significant improvements in the quality of generated text from both few-shot learning and fine-tuning perspectives using the PlanGTG dataset. Our study paves the way for new research directions in graph-to-text generation. PlanGTG datasets can be found in https://github.com/probe2/kg_text., Comment: NAACL 2025

Automated grading has become an essential tool in education technology due to its ability to efficiently assess large volumes of student work, provide consistent and unbiased evaluations, and deliver immediate feedback to enhance learning. However, current systems face significant limitations, including the need for large datasets in few-shot learning methods, a lack of personalized and actionable feedback, and an overemphasis on benchmark performance rather than student experience. To address these challenges, we propose a Zero-Shot Large Language Model (LLM)-Based Automated Assignment Grading (AAG) system. This framework leverages prompt engineering to evaluate both computational and explanatory student responses without requiring additional training or fine-tuning. The AAG system delivers tailored feedback that highlights individual strengths and areas for improvement, thereby enhancing student learning outcomes. Our study demonstrates the system's effectiveness through comprehensive evaluations, including survey responses from higher education students that indicate significant improvements in motivation, understanding, and preparedness compared to traditional grading methods. The results validate the AAG system's potential to transform educational assessment by prioritizing learning experiences and providing scalable, high-quality feedback.

I review the qualitative physical distinction between the Higgs and confinement phases of a gauge Higgs theory, and the non-local order parameter introduced by Matsuyama and myself which identifies the two phases. I then present some new results suggestive of a possible excitation spectrum of vector bosons in the electroweak sector of the Standard Model., Comment: Contribution to the Proceedings of the XVIth Quark Confinement and the Hadron Spectrum Conference (QCHSC24), Cairns, Australia, August 2024

Many real-world decision-making problems are combinatorial in nature, where states (e.g., surrounding traffic of a self-driving car) can be seen as a combination of basic elements (e.g., pedestrians, trees, and other cars). Due to combinatorial complexity, observing all combinations of basic elements in the training set is infeasible, which leads to an essential yet understudied problem of zero-shot generalization to states that are unseen combinations of previously seen elements. In this work, we first formalize this problem and then demonstrate how existing value-based reinforcement learning (RL) algorithms struggle due to unreliable value predictions in unseen states. We argue that this problem cannot be addressed with exploration alone, but requires more expressive and generalizable models. We demonstrate that behavior cloning with a conditioned diffusion model trained on expert trajectory generalizes better to states formed by new combinations of seen elements than traditional RL methods. Through experiments in maze, driving, and multiagent environments, we show that conditioned diffusion models outperform traditional RL techniques and highlight the broad applicability of our problem formulation.

We aim to understand physics of skill learning, i.e., how skills are learned in neural networks during training. We start by observing the Domino effect, i.e., skills are learned sequentially, and notably, some skills kick off learning right after others complete learning, similar to the sequential fall of domino cards. To understand the Domino effect and relevant behaviors of skill learning, we take physicists' approach of abstraction and simplification. We propose three models with varying complexities -- the Geometry model, the Resource model, and the Domino model, trading between reality and simplicity. The Domino effect can be reproduced in the Geometry model, whose resource interpretation inspires the Resource model, which can be further simplified to the Domino model. These models present different levels of abstraction and simplification; each is useful to study some aspects of skill learning. The Geometry model provides interesting insights into neural scaling laws and optimizers; the Resource model sheds light on the learning dynamics of compositional tasks; the Domino model reveals the benefits of modularity. These models are not only conceptually interesting -- e.g., we show how Chinchilla scaling laws can emerge from the Geometry model, but also are useful in practice by inspiring algorithmic development -- e.g., we show how simple algorithmic changes, motivated by these toy models, can speed up the training of deep learning models., Comment: 25 pages, 20 figures. Codes are available at https://github.com/KindXiaoming/physics_of_skill_learning

Large language models (LLMs) demonstrate remarkable performance, and improving their pre-training process appears to be key to enhancing their capabilities further. Based on the documented success of Adam, learning rate decay, and weight decay, we hypothesize that the pre-training loss landscape features a narrowing valley structure. Through experiments with synthetic loss functions, we discover that when gradient query noise is high relative to the valley's sharpness, Adam's performance falls behind that of Signum because Adam reduces the effective step size too drastically. This observation led us to develop FOCUS, an optimizer that enhances Signum by incorporating attraction toward moving averaged parameters, allowing it to handle noise better while maintaining larger step sizes. In training GPT-2, FOCUS proves to be more stable than Signum and faster than Adam. These results suggest that gradient noise may be an underappreciated limiting factor in LLM training, and FOCUS offers promising solutions., Comment: 19 pages, 8 figures

Hepatocellular carcinoma (HCC) is a leading cause of cancer-related mortality, with lung metastases being the most common site of distant spread and significantly worsening prognosis. Despite the growing availability of clinical and demographic data, predictive models for lung metastasis in HCC remain limited in scope and clinical applicability. In this study, we develop and validate an end-to-end machine learning pipeline using data from the Surveillance, Epidemiology, and End Results (SEER) database. We evaluated three machine learning models (Random Forest, XGBoost, and Logistic Regression) alongside a multilayer perceptron (MLP) neural network. Our models achieved high AUROC values and recall, with the Random Forest and MLP models demonstrating the best overall performance (AUROC = 0.82). However, the low precision across models highlights the challenges of accurately predicting positive cases. To address these limitations, we developed a custom loss function incorporating recall optimization, enabling the MLP model to achieve the highest sensitivity. An ensemble approach further improved overall recall by leveraging the strengths of individual models. Feature importance analysis revealed key predictors such as surgery status, tumor staging, and follow up duration, emphasizing the relevance of clinical interventions and disease progression in metastasis prediction. While this study demonstrates the potential of machine learning for identifying high-risk patients, limitations include reliance on imbalanced datasets, incomplete feature annotations, and the low precision of predictions. Future work should leverage the expanding SEER dataset, improve data imputation techniques, and explore advanced pre-trained models to enhance predictive accuracy and clinical utility., Comment: JJHK and GRN contributed equally, YD and TT are co-corresponding. 11 pages, 7 figures, 1 Table

The flexibility and complexity of IPv6 extension headers allow attackers to create covert channels or bypass security mechanisms, leading to potential data breaches or system compromises. The mature development of machine learning has become the primary detection technology option used to mitigate covert communication threats. However, the complexity of detecting covert communication, evolving injection techniques, and scarcity of data make building machine-learning models challenging. In previous related research, machine learning has shown good performance in detecting covert communications, but oversimplified attack scenario assumptions cannot represent the complexity of modern covert technologies and make it easier for machine learning models to detect covert communications. To bridge this gap, in this study, we analyzed the packet structure and network traffic behavior of IPv6, used encryption algorithms, and performed covert communication injection without changing network packet behavior to get closer to real attack scenarios. In addition to analyzing and injecting methods for covert communications, this study also uses comprehensive machine learning techniques to train the model proposed in this study to detect threats, including traditional decision trees such as random forests and gradient boosting, as well as complex neural network architectures such as CNNs and LSTMs, to achieve detection accuracy of over 90\%. This study details the methods used for dataset augmentation and the comparative performance of the applied models, reinforcing insights into the adaptability and resilience of the machine learning application in IPv6 covert communication. In addition, we also proposed a Generative AI-assisted interpretation concept based on prompt engineering as a preliminary study of the role of Generative AI agents in covert communication.

Generative modeling aims to generate new data samples that resemble a given dataset, with diffusion models recently becoming the most popular generative model. One of the main challenges of diffusion models is solving the problem in the input space, which tends to be very high-dimensional. Recently, solving diffusion models in the latent space through an encoder that maps from the data space to a lower-dimensional latent space has been considered to make the training process more efficient and has shown state-of-the-art results. The variational autoencoder (VAE) is the most commonly used encoder/decoder framework in this domain, known for its ability to learn latent representations and generate data samples. In this paper, we introduce a novel encoder/decoder framework with theoretical properties distinct from those of the VAE, specifically designed to preserve the geometric structure of the data distribution. We demonstrate the significant advantages of this geometry-preserving encoder in the training process of both the encoder and decoder. Additionally, we provide theoretical results proving convergence of the training process, including convergence guarantees for encoder training, and results showing faster convergence of decoder training when using the geometry-preserving encoder., Comment: 41 pages

This roadmap consolidates recent advances while exploring emerging applications, reflecting the remarkable diversity of hardware platforms, neuromorphic concepts, and implementation philosophies reported in the field. It emphasizes the critical role of cross-disciplinary collaboration in this rapidly evolving field.

ELIZA, created by Joseph Weizenbaum at MIT in the early 1960s, is usually considered the world's first chatbot. It was developed in MAD-SLIP on MIT's CTSS, the world's first time-sharing system, on an IBM 7094. We discovered an original ELIZA printout in Prof. Weizenbaum's archives at MIT, including an early version of the famous DOCTOR script, a nearly complete version of the MAD-SLIP code, and various support functions in MAD and FAP. Here we describe the reanimation of this original ELIZA on a restored CTSS, itself running on an emulated IBM 7094. The entire stack is open source, so that any user of a unix-like OS can run the world's first chatbot on the world's first time-sharing system., Comment: In review

At the risk of overstating the case, connectionist approaches to machine learning, i.e. neural networks, are enjoying a small vogue right now. However, these methods require large volumes of data and produce models that are uninterpretable to humans. An alternative framework that is compatible with neural networks and gradient-based learning, but explicitly models compositionality, is Vector Symbolic Architectures (VSAs). VSAs are a family of algebras on high-dimensional vector representations. They arose in cognitive science from the need to unify neural processing and the kind of symbolic reasoning that humans perform. While machine learning methods have benefited from category theoretical analyses, VSAs have not yet received similar treatment. In this paper, we present a first attempt at applying category theory to VSAs. Specifically, we conduct a brief literature survey demonstrating the lacking intersection of these two topics, provide a list of desiderata for VSAs, and propose that VSAs may be understood as a (division) rig in a category enriched over a monoid in Met (the category of Lawvere metric spaces). This final contribution suggests that VSAs may be generalised beyond current implementations. It is our hope that grounding VSAs in category theory will lead to more rigorous connections with other research, both within and beyond, learning and cognition., Comment: 13 pages, no figures, 2 tables, one appendix

This paper presents a novel machine-learning framework for reconstructing low-order gust-encounter flow field and lift coefficients from sparse, noisy surface pressure measurements. Our study thoroughly investigates the time-varying response of sensors to gust-airfoil interactions, uncovering valuable insights into optimal sensor placement. To address uncertainties in deep learning predictions, we implement probabilistic regression strategies to model both epistemic and aleatoric uncertainties. Epistemic uncertainty, reflecting the model's confidence in its predictions, is modeled using Monte Carlo dropout, as an approximation to the variational inference in the Bayesian framework, treating the neural network as a stochastic entity. On the other hand, aleatoric uncertainty, arising from noisy input measurements, is captured via learned statistical parameters, which propagates measurement noise through the network into the final predictions. Our results showcase the efficacy of this dual uncertainty quantification strategy in accurately predicting aerodynamic behavior under extreme conditions while maintaining computational efficiency, underscoring its potential to improve online sensor-based flow estimation in real-world applications.

In the quenched electroweak theory on the lattice I construct a set of physical states which overlap the physical photon and Z boson states. This is done by employing eigenstates of the covariant lattice Laplacian, in addition to the Higgs and lattice link variables, to construct gauge invariant vector boson creation operators. Diagonalizing the transfer matrix in the subspace of Hilbert space spanned by this set yields a massless photon and massive Z particle, as expected. But in the numerical data there is evidence for more vector bosons in the spectrum, albeit with considerable uncertainty in their masses, with the lowest finite mass particle in the range of 3-4 GeV., Comment: 7 pages, 8 figures

We pretrain METAGENE-1, a 7-billion-parameter autoregressive transformer model, which we refer to as a metagenomic foundation model, on a novel corpus of diverse metagenomic DNA and RNA sequences comprising over 1.5 trillion base pairs. This dataset is sourced from a large collection of human wastewater samples, processed and sequenced using deep metagenomic (next-generation) sequencing methods. Unlike genomic models that focus on individual genomes or curated sets of specific species, the aim of METAGENE-1 is to capture the full distribution of genomic information present within this wastewater, to aid in tasks relevant to pandemic monitoring and pathogen detection. We carry out byte-pair encoding (BPE) tokenization on our dataset, tailored for metagenomic sequences, and then pretrain our model. In this paper, we first detail the pretraining dataset, tokenization strategy, and model architecture, highlighting the considerations and design choices that enable the effective modeling of metagenomic data. We then show results of pretraining this model on our metagenomic dataset, providing details about our losses, system metrics, and training stability over the course of pretraining. Finally, we demonstrate the performance of METAGENE-1, which achieves state-of-the-art results on a set of genomic benchmarks and new evaluations focused on human-pathogen detection and genomic sequence embedding, showcasing its potential for public health applications in pandemic monitoring, biosurveillance, and early detection of emerging health threats.

The structural properties of packed soft-core particles provide a platform to understand the cross-pollinated physical concepts in solid-state- and soft-matter physics. Confined on spherical surface, the traditional differential geometry also dictates the overall defect properties in otherwise regular crystal lattices. Using molecular dynamics simulation of the Hertzian model as a tool, we report here the emergence of new types of disclination patterns: domain and counter-domain defects, when hexagonal and square patterns coexist. A new angle is presented to understand the incompatibility between tiling lattice shapes and the available spherical areal shapes, which is common in nature -- from molecular systems in biology to backbone construction in architectures., Comment: Authors' version of the article submitted to Nano Letters and accepted

Inspired by light-front holography, we compute the pion mass, charge radius, decay constant, electromagnetic form factor and electromagnetic transition form factor. To do so, we model the longitudinal quark dynamics using potentials due to 't Hooft and to Li & Vary. We find a longitudinal wavefunction that is rather more peaked about $x \sim 1/2$ than in previous studies. We also explore the strong degeneracy between these two potentials and conclude by noting that one scenario that accords well with the data also maps onto an equation previously noted by Vegh that describes the dynamics of a four-segmented string in $\mathrm{AdS}_3$., Comment: 22 pages, 5 figures

The increasing complexity of deep neural networks (DNNs) poses significant challenges for edge inference deployment due to resource and power constraints of edge devices. Recent works on unary-based matrix multiplication hardware aim to leverage data sparsity and low-precision values to enhance hardware efficiency. However, the adoption and integration of such unary hardware into commercial deep learning accelerators (DLA) remain limited due to processing element (PE) array dataflow differences. This work presents Tempus Core, a convolution core with highly scalable unary-based PE array comprising of tub (temporal-unary-binary) multipliers that seamlessly integrates with the NVDLA (NVIDIA's open-source DLA for accelerating CNNs) while maintaining dataflow compliance and boosting hardware efficiency. Analysis across various datapath granularities shows that for INT8 precision in 45nm CMOS, Tempus Core's PE cell unit (PCU) yields 59.3% and 15.3% reductions in area and power consumption, respectively, over NVDLA's CMAC unit. Considering a 16x16 PE array in Tempus Core, area and power improves by 75% and 62%, respectively, while delivering 5x and 4x iso-area throughput improvements for INT8 and INT4 precisions. Post-place and route analysis of Tempus Core's PCU shows that the 16x4 PE array for INT4 precision in 45nm CMOS requires only 0.017 mm^2 die area and consumes only 6.2mW of total power. We demonstrate that area-power efficient unary-based hardware can be seamlessly integrated into conventional DLAs, paving the path for efficient unary hardware for edge AI inference., Comment: Accepted in DATE 2025

Retrieval-augmented generation systems rely on effective document retrieval capabilities. By design, conventional sparse or dense retrievers face challenges in multi-hop retrieval scenarios. In this paper, we present GeAR, which advances RAG performance through two key innovations: (i) graph expansion, which enhances any conventional base retriever, such as BM25, and (ii) an agent framework that incorporates graph expansion. Our evaluation demonstrates GeAR's superior retrieval performance on three multi-hop question answering datasets. Additionally, our system achieves state-of-the-art results with improvements exceeding 10% on the challenging MuSiQue dataset, while requiring fewer tokens and iterations compared to other multi-step retrieval systems.

Artificial intelligence has advanced rapidly in the last decade, driven primarily by progress in the scale of deep-learning systems. Despite these advances, the creation of intelligent systems that can operate effectively in diverse, real-world environments remains a significant challenge. In this white paper, we outline the Thousand Brains Project, an ongoing research effort to develop an alternative, complementary form of AI, derived from the operating principles of the neocortex. We present an early version of a thousand-brains system, a sensorimotor agent that is uniquely suited to quickly learn a wide range of tasks and eventually implement any capabilities the human neocortex has. Core to its design is the use of a repeating computational unit, the learning module, modeled on the cortical columns found in mammalian brains. Each learning module operates as a semi-independent unit that can model entire objects, represents information through spatially structured reference frames, and both estimates and is able to effect movement in the world. Learning is a quick, associative process, similar to Hebbian learning in the brain, and leverages inductive biases around the spatial structure of the world to enable rapid and continual learning. Multiple learning modules can interact with one another both hierarchically and non-hierarchically via a "cortical messaging protocol" (CMP), creating more abstract representations and supporting multimodal integration. We outline the key principles motivating the design of thousand-brains systems and provide details about the implementation of Monty, our first instantiation of such a system. Code can be found at https://github.com/thousandbrainsproject/tbp.monty, along with more detailed documentation at https://thousandbrainsproject.readme.io/.

Retrieval augmentation, the practice of retrieving additional data from large auxiliary pools, has emerged as an effective technique for enhancing model performance in the low-data regime, e.g. few-shot learning. Prior approaches have employed only nearest-neighbor based strategies for data selection, which retrieve auxiliary samples with high similarity to instances in the target task. However, these approaches are prone to selecting highly redundant samples, since they fail to incorporate any notion of diversity. In our work, we first demonstrate that data selection strategies used in prior retrieval-augmented few-shot learning settings can be generalized using a class of functions known as Combinatorial Mutual Information (CMI) measures. We then propose COBRA (COmBinatorial Retrieval Augmentation), which employs an alternative CMI measure that considers both diversity and similarity to a target dataset. COBRA consistently outperforms previous retrieval approaches across image classification tasks and few-shot learning techniques when used to retrieve samples from LAION-2B. COBRA introduces negligible computational overhead to the cost of retrieval while providing significant gains in downstream model performance.

Large language models (LLMs) have demonstrated impressive capabilities in various reasoning tasks but face significant challenges with complex, knowledge-intensive multi-hop queries, particularly those involving new or long-tail knowledge. Existing benchmarks often fail to fully address these challenges. To bridge this gap, we introduce MINTQA (Multi-hop Question Answering on New and Tail Knowledge), a comprehensive benchmark to evaluate LLMs' capabilities in multi-hop reasoning across four critical dimensions: question handling strategy, sub-question generation, retrieval-augmented generation, and iterative or dynamic decomposition and retrieval. MINTQA comprises 10,479 question-answer pairs for evaluating new knowledge and 17,887 pairs for assessing long-tail knowledge, with each question equipped with corresponding sub-questions and answers. Our systematic evaluation of 22 state-of-the-art LLMs on MINTQA reveals significant limitations in their ability to handle complex knowledge base queries, particularly in handling new or unpopular knowledge. Our findings highlight critical challenges and offer insights for advancing multi-hop reasoning capabilities. The MINTQA benchmark is available at https://github.com/probe2/multi-hop/.

Binaries harboring a millisecond pulsar (MSP) and a black hole (BH) are a key observing target for current and upcoming pulsar surveys. We model the formation and evolution of such binaries in isolation at solar metallicity using the next-generation binary population synthesis code POSYDON. We examine neutron star (NS)-BH binaries where the NS forms first (labeled NSBH), as the NS must be able to spin-up to MSP rotation periods before the BH forms in these systems. We find that NSBHs are very rare and have a birth rate < 1 Myr$^{-1}$ for a Milky Way-like galaxy in our typical models. The NSBH birth rate is 2-3 orders of magnitude smaller than that for NS-BHs where the BH forms first (labeled BHNS). These rates are also sensitive to model assumptions about the supernova (SN) remnant masses, natal kicks, and common-envelope efficiency. We find that 100% of NSBHs undergo a mass ratio reversal before the first SN and up to 64% of NSBHs undergo a double common envelope phase after the mass ratio reversal occurs. Most importantly, no NSBH binaries in our populations undergo a mass transfer phase, either stable or unstable, after the first SN. This implies that there is no possibility of pulsar spin-up via accretion, and thus MSP-BH binaries cannot form. Thus, dynamical environments and processes may provide the only formation channels for such MSP-BH binaries., Comment: Submitted to ApJ, 16 pages, 6 figures, 1 table

Dynamic and adaptive mesh refinement is pivotal in high-resolution, multi-physics, multi-model simulations, necessitating precise physics resolution in localized areas across expansive domains. Today's supercomputers' extreme heterogeneity presents a significant challenge for dynamically adaptive codes, highlighting the importance of achieving performance portability at scale. Our research focuses on astrophysical simulations, particularly stellar mergers, to elucidate early universe dynamics. We present Octo-Tiger, leveraging Kokkos, HPX, and SIMD for portable performance at scale in complex, massively parallel adaptive multi-physics simulations. Octo-Tiger supports diverse processors, accelerators, and network backends. Experiments demonstrate exceptional scalability across several heterogeneous supercomputers including Perlmutter, Frontier, and Fugaku, encompassing major GPU architectures and x86, ARM, and RISC-V CPUs. Parallel efficiency of 47.59% (110,080 cores and 6880 hybrid A100 GPUs) on a full-system run on Perlmutter (26% HPCG peak performance) and 51.37% (using 32,768 cores and 2,048 MI250X) on Frontier are achieved.

Cosmologically stable, light particles that came into thermal contact with the Standard Model in the early universe may persist today as a form of hot dark matter. For relics with masses in the eV range, their role in structure formation depends critically on their mass. We trace the evolution of such hot relics and derive their density profiles around cold dark matter halos, introducing a framework for their indirect detection. Applying this framework to axions -- a natural candidate for a particle that can reach thermal equilibrium with the Standard Model in the early universe and capable of decaying into two photons -- we establish stringent limits on the axion-photon coupling $g_{a \gamma} $ using current observations of dwarf galaxies, the Milky Way halo, and galaxy clusters. Our results set new bounds on hot axions in the $\mathcal{O}(1-10)\,$eV range., Comment: 30 pages, 6 figures

In response to the call for agent-based solutions that leverage the ever-increasing capabilities of the deep models' ecosystem, we introduce Hive -- a comprehensive solution for selecting appropriate models and subsequently planning a set of atomic actions to satisfy the end-users' instructions. Hive operates over sets of models and, upon receiving natural language instructions (i.e. user queries), schedules and executes explainable plans of atomic actions. These actions can involve one or more of the available models to achieve the overall task, while respecting end-users specific constraints. Notably, Hive handles tasks that involve multi-modal inputs and outputs, enabling it to handle complex, real-world queries. Our system is capable of planning complex chains of actions while guaranteeing explainability, using an LLM-based formal logic backbone empowered by PDDL operations. We introduce the MuSE benchmark in order to offer a comprehensive evaluation of the multi-modal capabilities of agent systems. Our findings show that our framework redefines the state-of-the-art for task selection, outperforming other competing systems that plan operations across multiple models while offering transparency guarantees while fully adhering to user constraints., Comment: Under review

Humans distill complex experiences into fundamental abstractions that enable rapid learning and adaptation. Similarly, autoregressive transformers exhibit adaptive learning through in-context learning (ICL), which begs the question of how. In this paper, we propose concept encoding-decoding mechanism to explain ICL by studying how transformers form and use internal abstractions in their representations. On synthetic ICL tasks, we analyze the training dynamics of a small transformer and report the coupled emergence of concept encoding and decoding. As the model learns to encode different latent concepts (e.g., ``Finding the first noun in a sentence.") into distinct, separable representations, it concureently builds conditional decoding algorithms and improve its ICL performance. We validate the existence of this mechanism across pretrained models of varying scales (Gemma-2 2B/9B/27B, Llama-3.1 8B/70B). Further, through mechanistic interventions and controlled finetuning, we demonstrate that the quality of concept encoding is causally related and predictive of ICL performance. Our empirical insights shed light into better understanding the success and failure modes of large language models via their representations.

To review the lessons learnt from recent deep geothermal case studies and plan strategically the research, development, regulation, and communication work required for the implementation of an Enhanced Geothermal System (EGS) at Cornell University, a group of engineers and scholars convened a two-day workshop on the Ithaca campus, on October 23-24, 2024. The event was funded by Cornell Atkinson Center for Sustainability. This report is a summary of the content of the presentations and discussions that took place during the workshop. The first section focuses on philosophical, sociological, economic, and regulatory questions posed by EGS deployment as a means to mitigate climate change. The second section tackles the scientific and technological research areas associated with EGS. The third section aims to assess the feasibility of developing EGS for heat direct use at Cornell University, based on results and information available to date. The report concludes with a summary of the most salient technological and scientific breakthroughs, and a plan for future technological and academic engagement in EGS projects at Cornell.

We present a broadband and polarization-insensitive unidirectional imager that operates at the visible part of the spectrum, where image formation occurs in one direction while in the opposite direction, it is blocked. This approach is enabled by deep learning-driven diffractive optical design with wafer-scale nano-fabrication using high-purity fused silica to ensure optical transparency and thermal stability. Our design achieves unidirectional imaging across three visible wavelengths (covering red, green and blue parts of the spectrum), and we experimentally validated this broadband unidirectional imager by creating high-fidelity images in the forward direction and generating weak, distorted output patterns in the backward direction, in alignment with our numerical simulations. This work demonstrates the wafer-scale production of diffractive optical processors, featuring 16 levels of nanoscale phase features distributed across two axially aligned diffractive layers for visible unidirectional imaging. This approach facilitates mass-scale production of ~0.5 billion nanoscale phase features per wafer, supporting high-throughput manufacturing of hundreds to thousands of multi-layer diffractive processors suitable for large apertures and parallel processing of multiple tasks. Our design can seamlessly integrate into conventional optical systems, broadening its applicability in fields such as security, defense, and telecommunication. Beyond broadband unidirectional imaging in the visible spectrum, this study establishes a pathway for artificial-intelligence-enabled diffractive optics with versatile applications, signaling a new era in optical device functionality with industrial-level massively scalable fabrication., Comment: 23 Pages, 6 Figures

Cardiac Magnetic Resonance (CMR) imaging is widely used for heart modelling and digital twin computational analysis due to its ability to visualize soft tissues and capture dynamic functions. However, the anisotropic nature of CMR images, characterized by large inter-slice distances and misalignments from cardiac motion, poses significant challenges to accurate model reconstruction. These limitations result in data loss and measurement inaccuracies, hindering the capture of detailed anatomical structures. This study introduces MorphiNet, a novel network that enhances heart model reconstruction by leveraging high-resolution Computer Tomography (CT) images, unpaired with CMR images, to learn heart anatomy. MorphiNet encodes anatomical structures as gradient fields, transforming template meshes into patient-specific geometries. A multi-layer graph subdivision network refines these geometries while maintaining dense point correspondence. The proposed method achieves high anatomy fidelity, demonstrating approximately 40% higher Dice scores, half the Hausdorff distance, and around 3 mm average surface error compared to state-of-the-art methods. MorphiNet delivers superior results with greater inference efficiency. This approach represents a significant advancement in addressing the challenges of CMR-based heart model reconstruction, potentially improving digital twin computational analyses of cardiac structure and functions.

Multimodal entity linking (MEL) aims to link ambiguous mentions within multimodal contexts to corresponding entities in a multimodal knowledge base. Most existing approaches to MEL are based on representation learning or vision-and-language pre-training mechanisms for exploring the complementary effect among multiple modalities. However, these methods suffer from two limitations. On the one hand, they overlook the possibility of considering negative samples from the same modality. On the other hand, they lack mechanisms to capture bidirectional cross-modal interaction. To address these issues, we propose a Multi-level Matching network for Multimodal Entity Linking (M3EL). Specifically, M3EL is composed of three different modules: (i) a Multimodal Feature Extraction module, which extracts modality-specific representations with a multimodal encoder and introduces an intra-modal contrastive learning sub-module to obtain better discriminative embeddings based on uni-modal differences; (ii) an Intra-modal Matching Network module, which contains two levels of matching granularity: Coarse-grained Global-to-Global and Fine-grained Global-to-Local, to achieve local and global level intra-modal interaction; (iii) a Cross-modal Matching Network module, which applies bidirectional strategies, Textual-to-Visual and Visual-to-Textual matching, to implement bidirectional cross-modal interaction. Extensive experiments conducted on WikiMEL, RichpediaMEL, and WikiDiverse datasets demonstrate the outstanding performance of M3EL when compared to the state-of-the-art baselines., Comment: Accepted at KDD'25

Standard deep learning architectures used for classification generate label predictions with a projection head and softmax activation function. Although successful, these methods fail to leverage the relational information between samples in the batch for generating label predictions. In recent works, graph-based learning techniques, namely Laplace learning, have been heuristically combined with neural networks for both supervised and semi-supervised learning (SSL) tasks. However, prior works approximate the gradient of the loss function with respect to the graph learning algorithm or decouple the processes; end-to-end integration with neural networks is not achieved. In this work, we derive backpropagation equations, via the adjoint method, for inclusion of a general family of graph learning layers into a neural network. This allows us to precisely integrate graph Laplacian-based label propagation into a neural network layer, replacing a projection head and softmax activation function for classification tasks. Using this new framework, our experimental results demonstrate smooth label transitions across data, improved robustness to adversarial attacks, improved generalization, and improved training dynamics compared to the standard softmax-based approach., Comment: 44 pages, 11 figures. Preprint. Submitted to the Journal of Machine Learning Research

Robust ranking and selection (R&S) is an important and challenging variation of conventional R&S that seeks to select the best alternative among a finite set of alternatives. It captures the common input uncertainty in the simulation model by using an ambiguity set to include multiple possible input distributions and shifts to select the best alternative with the smallest worst-case mean performance over the ambiguity set. In this paper, we aim at developing new fixed-budget robust R&S procedures to minimize the probability of incorrect selection (PICS) under a limited sampling budget. Inspired by an additive upper bound of the PICS, we derive a new asymptotically optimal solution to the budget allocation problem. Accordingly, we design a new sequential optimal computing budget allocation (OCBA) procedure to solve robust R&S problems efficiently. We then conduct a comprehensive numerical study to verify the superiority of our robust OCBA procedure over existing ones. The numerical study also provides insights on the budget allocation behaviors that lead to enhanced efficiency.

We investigate the charging dynamics of Sachdev-Ye-Kitaev (SYK) models as quantum batteries, highlighting their capacity to achieve quantum charging advantages. By analytically deriving the scaling of the charging power in SYK batteries, we identify the two key mechanisms underlying this advantage: the use of operators scaling extensively with system size $N$ and the facilitation of operator delocalization by specific graph structures. A novel graph-theoretic framework is introduced in which the charging process is recast as a random walk on a graph, enabling a quantitative analysis of operator spreading. Our results establish rigorous conditions for the quantum advantage in SYK batteries and extend these insights to graph-based SYK models, revealing broader implications for energy storage and quantum dynamics. This work opens avenues for leveraging quantum chaos and complex network structures in optimizing energy transfer processes., Comment: 9 pages, 4 figures

High performance computing (HPC) architectures have undergone rapid development in recent years. As a result, established software suites face an ever increasing challenge to remain performant on and portable across modern systems. Many of the widely adopted atmospheric modeling codes cannot fully (or in some cases, at all) leverage the acceleration provided by General-Purpose Graphics Processing Units (GPGPUs), leaving users of those codes constrained to increasingly limited HPC resources. Energy Research and Forecasting (ERF) is a regional atmospheric modeling code that leverages the latest HPC architectures, whether composed of only Central Processing Units (CPUs) or incorporating GPUs. ERF contains many of the standard discretizations and basic features needed to model general atmospheric dynamics as well as flows relevant to renewable energy. The modular design of ERF provides a flexible platform for exploring different physics parameterizations and numerical strategies. ERF is built on a state-of-the-art, well-supported, software framework (AMReX) that provides a performance portable interface and ensures ERF's long-term sustainability on next generation computing systems. This paper details the numerical methodology of ERF and presents results for a series of verification and validation cases.

Water has proven to be ubiquitously detected in near-infrared (NIR) transmission spectroscopy observations of hot Jupiter atmospheres, including WASP-17b. However, previous analyses of WASP-17b's atmosphere based upon Hubble Space Telescope (HST) and Spitzer data could not constrain the water abundance, finding that sub-solar, super-solar and bimodal posterior distributions were all statistically valid. In this work, we observe one transit of the hot Jupiter WASP-17b using JWST's Near Infrared Imager and Slitless Spectrograph Single Object Slitless Spectroscopy (NIRISS SOSS) mode. We analyze our data using three independent data analysis pipelines, finding excellent agreement between results. Our transmission spectrum shows multiple H$_2$O absorption features and a flatter slope towards the optical than seen in previous HST observations. We analyze our spectrum using both PICASO+Virga forward models and free retrievals. POSEIDON retrievals provide a well-constrained super-solar $\log$(H$_2$O) abundance (-2.96$^{+0.31}_{-0.24}$), breaking the degeneracy from the previous HST/Spitzer analysis. We verify our POSEIDON results with petitRADTRANS retrievals. Additionally, we constrain the abundance of $\log$(H$^-$), -10.19$^{+0.30}_{-0.23}$, finding that our model including H$^-$ is preferred over our model without H$^-$ to 5.1 $\sigma$. Furthermore, we constrain the $\log$(K) abundance (-8.07$^{+0.58}_{-0.52}$) in WASP-17b's atmosphere for the first time using space-based observations. Our abundance constraints demonstrate the power of NIRISS SOSS's increased resolution, precision, and wavelength range to improve upon previous NIR space-based results. This work is part of a series of studies by our JWST Telescope Scientist Team (JWST-TST), in which we use Guaranteed Time Observations to perform Deep Reconnaissance of Exoplanet Atmospheres through Multi-instrument Spectroscopy (DREAMS)., Comment: Accepted for Publication in the Astronomical Journal (AJ) on 19 November 2024; 37 Pages, 17 Figures, 7 Tables. This version adds 2 co-authors, updates affiliations/acknowledgements, and corrects minor typos. Zenodo link to data is at https://zenodo.org/records/14193061

Computer-aided synthesis planning (CASP) has made significant strides in generating retrosynthetic pathways for simple molecules in a non-constrained fashion. Recent work introduces a specialised bidirectional search algorithm with forward and retro expansion to address the starting material-constrained synthesis problem, allowing CASP systems to provide synthesis pathways from specified starting materials, such as waste products or renewable feed-stocks. In this work, we introduce a simple guided search which allows solving the starting material-constrained synthesis planning problem using an existing, uni-directional search algorithm, Retro*. We show that by optimising a single hyperparameter, Tango* outperforms existing methods in terms of efficiency and solve rate. We find the Tango* cost function catalyses strong improvements for the bidirectional DESP methods. Our method also achieves lower wall clock times while proposing synthetic routes of similar length, a common metric for route quality. Finally, we highlight potential reasons for the strong performance of Tango over neural guided search methods

Artificial Intelligence (AI), like any transformative technology, has the potential to be a double-edged sword, leading either toward significant advancements or detrimental outcomes for society as a whole. As is often the case when it comes to widely-used technologies in market economies (e.g., cars and semiconductor chips), commercial interest tends to be the predominant guiding factor. The AI community is at risk of becoming polarized to either take a laissez-faire attitude toward AI development, or to call for government overregulation. Between these two poles we argue for the community of AI practitioners to consciously and proactively work for the common good. This paper offers a blueprint for a new type of innovation infrastructure including 18 concrete milestones to guide AI research in that direction. Our view is that we are still in the early days of practical AI, and focused efforts by practitioners, policymakers, and other stakeholders can still maximize the upsides of AI and minimize its downsides. We talked to luminaries such as recent Nobelist John Jumper on science, President Barack Obama on governance, former UN Ambassador and former National Security Advisor Susan Rice on security, philanthropist Eric Schmidt on several topics, and science fiction novelist Neal Stephenson on entertainment. This ongoing dialogue and collaborative effort has produced a comprehensive, realistic view of what the actual impact of AI could be, from a diverse assembly of thinkers with deep understanding of this technology and these domains. From these exchanges, five recurring guidelines emerged, which form the cornerstone of a framework for beginning to harness AI in service of the public good. They not only guide our efforts in discovery but also shape our approach to deploying this transformative technology responsibly and ethically.

We present the MULTIMODAL UNIVERSE, a large-scale multimodal dataset of scientific astronomical data, compiled specifically to facilitate machine learning research. Overall, the MULTIMODAL UNIVERSE contains hundreds of millions of astronomical observations, constituting 100\,TB of multi-channel and hyper-spectral images, spectra, multivariate time series, as well as a wide variety of associated scientific measurements and "metadata". In addition, we include a range of benchmark tasks representative of standard practices for machine learning methods in astrophysics. This massive dataset will enable the development of large multi-modal models specifically targeted towards scientific applications. All codes used to compile the MULTIMODAL UNIVERSE and a description of how to access the data is available at https://github.com/MultimodalUniverse/MultimodalUniverse, Comment: Accepted at NeurIPS Datasets and Benchmarks track

We show that a restricted version of a conjecture of M. Talagrand on the relation between "expectation thresholds" and "fractional expectation thresholds" follows easily from a strong version of a second conjecture of Talagrand, on "selector processes." The selector process conjecture was proved by Park and Pham, and the quantitative strengthening used here is due to Bednorz, Martynek, and Meller., Comment: 3 pages

During the COVID-19 pandemic there were many changes in arts and design alumni's paid work, creative work, income and in the skills that they needed to be successful in their careers. Given the documented impact of the pandemic on the arts and culture sector, on arts and creative workers, and on institutions of higher education, the authors turn to data from the 2022 Strategic National Arts Alumni Project (SNAAP) survey to understand how the impact of the pandemic was felt by people doing arts, design, or creative work as part of their job or occupational roles and the varied experiences across socio-demographic characteristics. This report focuses on providing insight into arts and design graduates whose work is--or before the pandemic was--in the arts, design, or creative realm by examining their working lives, economic lives, and the skills that they needed to attain or update during the first two years of the pandemic in the United States. In addition to presenting analyses and findings about arts and design alumni's experiences of the first year of the pandemic along these dimensions of their creative and working lives, the authors also dedicate a section of the conclusion of this report to the impressions and challenges identified by higher education arts leaders in a series of formal conversation at the beginning of 2021 to frame commentary about university preparation for resilient arts and design alumni. [Additional funding for this report was provided by the University of Illinois College of Fine and Applied Arts, the University of Illinois Investment for Growth Fund, and the Department of Arts Administration, Education, and Policy at The Ohio State University.]