Search

Motivated by the experimental identification of magnetic compounds consisting of zigzag chains, we analyze the band structure topology of magnons in ferromagnets on a zigzag lattice. We account for the general lattice geometry by including spatially anisotropic Heisenberg exchange interactions and by Dzyaloshinskii-Moriya interaction on inversion asymmetric bonds. Within the linear spin-wave theory, we find two magnon branches, whose band structure topology (i.e., Chern numbers) we map out in a comprehensive phase diagram. Notably, besides topologically trivial and gapless phases, we identify topologically nontrivial phases that support chiral edge magnons. We show that these edge states are robust against elastic defect scattering., Comment: 8 pages, 7 figures

For noncovalent interactions (NCIs), the CCSD(T) coupled cluster method is widely regarded as the `gold standard'. With localized orbital approximations, benchmarks for ever larger NCI complexes are being published; yet tantalizing evidence from quantum Monte Carlo (QMC) results appears to indicate that as the system size grows, CCSD(T) overbinds NCIs by progressively larger amounts, particularly when $\pi$-stacking is involved. Alas, post-CCSD(T) methods like CCSDT(Q) are cost-prohibitive, which requires us to consider alternative means of estimating post-CCSD(T) contributions. In this work, we take a step back by considering the evolution of the correlation energy with respect to the number of subunits for such $\pi$-stacked sequences as acene dimers and alkadiene dimers. We show it to be almost perfectly linear, and propose the slope of the line as a probe for the behavior of a given electron correlation method. By comparison with rank-reduced CCSDT(Q) results for benzene and naphthalene dimers, we show that while CCSD(T) does slightly overbind, it does not at the level suggested by the QMC results., Comment: 31 pages, 3 figures; 2nd updated version

Fairness metrics are a core tool in the fair machine learning literature (FairML), used to determine that ML models are, in some sense, "fair". Real-world data, however, are typically plagued by various measurement biases and other violated assumptions, which can render fairness assessments meaningless. We adapt tools from causal sensitivity analysis to the FairML context, providing a general framework which (1) accommodates effectively any combination of fairness metric and bias that can be posed in the "oblivious setting"; (2) allows researchers to investigate combinations of biases, resulting in non-linear sensitivity; and (3) enables flexible encoding of domain-specific constraints and assumptions. Employing this framework, we analyze the sensitivity of the most common parity metrics under 3 varieties of classifier across 14 canonical fairness datasets. Our analysis reveals the striking fragility of fairness assessments to even minor dataset biases. We show that causal sensitivity analysis provides a powerful and necessary toolkit for gauging the informativeness of parity metric evaluations. Our repository is available here: https://github.com/Jakefawkes/fragile_fair., Comment: Published at Neurips 2024 in the Dataset and Benchmarks Track

Quantization has established itself as the primary approach for decreasing the computational and storage expenses associated with Large Language Models (LLMs) inference. The majority of current research emphasizes quantizing weights and activations to enable low-bit general-matrix-multiply (GEMM) operations, with the remaining non-linear operations executed at higher precision. In our study, we discovered that following the application of these techniques, the primary bottleneck in LLMs inference lies in the softmax layer. The softmax operation comprises three phases: exponent calculation, accumulation, and normalization, Our work focuses on optimizing the first two phases. We propose an analytical approach to determine the optimal clipping value for the input to the softmax function, enabling sub-4-bit quantization for LLMs inference. This method accelerates the calculations of both $e^x$ and $\sum(e^x)$ with minimal to no accuracy degradation. For example, in LLaMA1-30B, we achieve baseline performance with 2-bit quantization on the well-known "Physical Interaction: Question Answering" (PIQA) dataset evaluation. This ultra-low bit quantization allows, for the first time, an acceleration of approximately 4x in the accumulation phase. The combination of accelerating both $e^x$ and $\sum(e^x)$ results in a 36.9% acceleration in the softmax operation.

Structural understanding of complex visual objects is an important unsolved component of artificial intelligence. To study this, we develop a new technique for the recently proposed Break-and-Make problem in LTRON where an agent must learn to build a previously unseen LEGO assembly using a single interactive session to gather information about its components and their structure. We attack this problem by building an agent that we call \textbf{\ours} that is able to make its own visual instruction book. By disassembling an unseen assembly and periodically saving images of it, the agent is able to create a set of instructions so that it has the information necessary to rebuild it. These instructions form an explicit memory that allows the model to reason about the assembly process one step at a time, avoiding the need for long-term implicit memory. This in turn allows us to train on much larger LEGO assemblies than has been possible in the past. To demonstrate the power of this model, we release a new dataset of procedurally built LEGO vehicles that contain an average of 31 bricks each and require over one hundred steps to disassemble and reassemble. We train these models using online imitation learning which allows the model to learn from its own mistakes. Finally, we also provide some small improvements to LTRON and the Break-and-Make problem that simplify the learning environment and improve usability.

Segmentation is a crucial step in microscopy image analysis. Numerous approaches have been developed over the past years, ranging from classical segmentation algorithms to advanced deep learning models. While U-Net remains one of the most popular and well-established models for biomedical segmentation tasks, recently developed transformer-based models promise to enhance the segmentation process of microscopy images. In this work, we assess the efficacy of transformers, including UNETR, the Segment Anything Model, and Swin-UPerNet, and compare them with the well-established U-Net model across various image modalities such as electron microscopy, brightfield, histopathology, and phase-contrast. Our evaluation identifies several limitations in the original Swin Transformer model, which we address through architectural modifications to optimise its performance. The results demonstrate that these modifications improve segmentation performance compared to the classical U-Net model and the unmodified Swin-UPerNet. This comparative analysis highlights the promise of transformer models for advancing biomedical image segmentation. It demonstrates that their efficiency and applicability can be improved with careful modifications, facilitating their future use in microscopy image analysis tools., Comment: to be published in ECCV 2024 BioImage Computing Workshop

We train, for the first time, large language models using FP8 precision on datasets up to 2 trillion tokens -- a 20-fold increase over previous limits. Through these extended training runs, we uncover critical instabilities in FP8 training that were not observable in earlier works with shorter durations. We trace these instabilities to outlier amplification by the SwiGLU activation function. Interestingly, we show, both analytically and empirically, that this amplification happens only over prolonged training periods, and link it to a SwiGLU weight alignment process. To address this newly identified issue, we introduce Smooth-SwiGLU, a novel modification that ensures stable FP8 training without altering function behavior. We also demonstrate, for the first time, FP8 quantization of both Adam optimizer moments. Combining these innovations, we successfully train a 7B parameter model using FP8 precision on 256 Intel Gaudi2 accelerators, achieving on-par results with the BF16 baseline while delivering up to a $\sim 34 \%$ throughput improvement.

We study the Berry curvature and Chern number of a non-collinear spin state on a honeycomb lattice that evolves from coplanar to ferromagnetic with a magnetic field applied along the $z$ axis. The coplanar state is stabilized by nearest-neighbor ferromagnetic interactions, single-ion anisotropy along $z$, and Dzyalloshinskii-Moriya interactions between next-nearest neighbor sites. Below the critical field $H_c$ that aligns the spins, the magnetic unit cell contains $M=6$ sites and the spin dynamics contains six magnon subbands. Although the classical energy is degenerate wrt the twist angle $\phi $ between nearest-neighbor spins, the dependence of the free energy on $\phi $ at low temperatures is dominated by the magnon zero-point energy, which contains extremum at $\phi =\pi l/3$ for integer $l$. The only unique ground states GS($\phi )$ have $l=0$ or 1. For $H < H_c'$, the zero-point energy has minima at even $l$ and the ground state is GS(0). For $H_c' < H < H_c$, the zero-point energy has minima at odd $l$ and the ground state is GS($\pi/3$). In GS(0), the magnon density-of-states exhibits five distinct phases with increasing field associated with the opening and closing of energy gaps between the two or three magnonic bands, each containing between 1 and 4 four magnon subbands. While the Berry curvature vanishes for the coplanar $\phi=0$ phase in zero field, the Berry curvature and Chern numbers exhibit signatures of the five phases at nonzero fields below $H_c'$. If $\phi \ne \pi l/3$, the Chern numbers of the two or three magnonic bands are non-integer. We also evaluate the inelastic neutron-scattering spectrum $S(\vk ,\omega )$ produced by the six magnon subbands in all five phases of GS(0) and in GS($\pi/3$)., Comment: 13 pages, 12 figures

We have investigated the title question for both a subset of the W4-11 total atomization energies benchmark, and for the A24x8 noncovalent interactions benchmark. Overall, counterpoise corrections to post-CCSD(T) contributions are about two orders of magnitude less important than those to the CCSD(T) interaction energy. Counterpoise corrections for connected quadruple substitutions (Q) are negligible, and $(Q)_\Lambda - (Q)$ or $T_4 - (Q)$ especially so. In contrast, for atomization energies, the $T_3-(T)$ counterpoise correction can reach about 0.05 \kcalmol~for small basis sets like cc-pVDZ, thought it rapidly tapers off with cc-pVTZ and especially aug-cc-pVTZ basis sets. It is reduced to insignificance by the extrapolation of $T_3-(T)$ applied in both W4 and HEAT thermochemistry protocols. In noncovalent dimers, the differential BSSE on post-CCSD(T) correlation contributions is negligible even in basis sets as small as the unpolarized split-valence cc-pVDZ(no d)., Comment: Final published article under CC:BY license (Hans Lischka festschrift)

Large language models (LLMs) have recently seen widespread adoption, in both academia and industry. As these models grow, they become valuable intellectual property (IP), reflecting enormous investments by their owners. Moreover, the high cost of cloud-based deployment has driven interest towards deployment to edge devices, yet this risks exposing valuable parameters to theft and unauthorized use. Current methods to protect models' IP on the edge have limitations in terms of practicality, loss in accuracy, or suitability to requirements. In this paper, we introduce a novel hybrid inference algorithm, named SLIP, designed to protect edge-deployed models from theft. SLIP is the first hybrid protocol that is both practical for real-world applications and provably secure, while having zero accuracy degradation and minimal impact on latency. It involves partitioning the model between two computing resources, one secure but expensive, and another cost-effective but vulnerable. This is achieved through matrix decomposition, ensuring that the secure resource retains a maximally sensitive portion of the model's IP while performing a minimal amount of computations, and vice versa for the vulnerable resource. Importantly, the protocol includes security guarantees that prevent attackers from exploiting the partition to infer the secured information. Finally, we present experimental results that show the robustness and effectiveness of our method, positioning it as a compelling solution for protecting LLMs.

We propose a robot learning method for communicating, planning, and executing a wide range of tasks, dubbed This&That. We achieve robot planning for general tasks by leveraging the power of video generative models trained on internet-scale data containing rich physical and semantic context. In this work, we tackle three fundamental challenges in video-based planning: 1) unambiguous task communication with simple human instructions, 2) controllable video generation that respects user intents, and 3) translating visual planning into robot actions. We propose language-gesture conditioning to generate videos, which is both simpler and clearer than existing language-only methods, especially in complex and uncertain environments. We then suggest a behavioral cloning design that seamlessly incorporates the video plans. This&That demonstrates state-of-the-art effectiveness in addressing the above three challenges, and justifies the use of video generation as an intermediate representation for generalizable task planning and execution. Project website: https://cfeng16.github.io/this-and-that/.

Tensor network contractions are widely used in statistical physics, quantum computing, and computer science. We introduce a method to efficiently approximate tensor network contractions using low-rank approximations, where each intermediate tensor generated during the contractions is approximated as a low-rank binary tree tensor network. The proposed algorithm has the flexibility to incorporate a large portion of the environment when performing low-rank approximations, which can lead to high accuracy for a given rank. Here, the environment refers to the remaining set of tensors in the network, and low-rank approximations with larger environments can generally provide higher accuracy. For contracting tensor networks defined on lattices, the proposed algorithm can be viewed as a generalization of the standard boundary-based algorithms. In addition, the algorithm includes a cost-efficient density matrix algorithm for approximating a tensor network with a general graph structure into a tree structure, whose computational cost is asymptotically upper-bounded by that of the standard algorithm that uses canonicalization. Experimental results indicate that the proposed technique outperforms previously proposed approximate tensor network contraction algorithms for multiple problems in terms of both accuracy and efficiency.

The design and development of power electronics converters pose a multitude of challenges. The evaluation of power electronics converters, particularly when operating at high power levels, presents a significant task, offering designers a deeper understanding of the functionality. Several methodologies have been devised to conduct hardware-in-the-loop (HIL) tests, which are classified into two main categories: controller hardware-in-the-loop (CHIL) and power hardware-in-the-loop (PHIL) tests. This paper explores the advantages and drawbacks of these two approaches and introduces a straightforward and cost-effective CHIL method for initial proof of concept and risk-free controller training. This method is based on the interaction between MATLAB/Simulink and Python. To assess the operation of the proposed system, a modeled battery pack (96S1P) is charged using a modeled battery charger converter, employing a non-linear PID controller. In this scenario, the controller benefits from the CC-CV technique for charging the battery pack. The results are presented in the final section., Comment: 2024 IEEE Transportation Electrification Conference & Expo (ITEC 2024) (5 pages, 11 figures)

Basis set extrapolations are typically rationalized either from analytical arguments involving the partial-wave or principal expansions of the correlation energy in helium-like systems, or from fitting extrapolation parameters to reference energetics for a small(ish) training set. Seeking to avoid both, we explore a third alternative: extracting extrapolation parameters from the requirement that the BSSE (basis set superposition error) should vanish at the complete basis set limit. We find this to be a viable approach provided that the underlying basis sets are not too small and reasonably well balanced. For basis sets not augmented by diffuse functions, BSSE minimization and energy fitting yield quite similar parameters., Comment: J. Phys. Chem. A, ASAP (Gustavo E. Scuseria festschrift); CC:BY 4.0 Open Access

Deep learning is dramatically transforming the field of medical imaging and radiology, enabling the identification of pathologies in medical images, including computed tomography (CT) and X-ray scans. However, the performance of deep learning models, particularly in segmentation tasks, is often limited by the need for extensive annotated datasets. To address this challenge, the capabilities of weakly supervised semantic segmentation are explored through the lens of Explainable AI and the generation of counterfactual explanations. The scope of this research is development of a novel counterfactual inpainting approach (COIN) that flips the predicted classification label from abnormal to normal by using a generative model. For instance, if the classifier deems an input medical image X as abnormal, indicating the presence of a pathology, the generative model aims to inpaint the abnormal region, thus reversing the classifier's original prediction label. The approach enables us to produce precise segmentations for pathologies without depending on pre-existing segmentation masks. Crucially, image-level labels are utilized, which are substantially easier to acquire than creating detailed segmentation masks. The effectiveness of the method is demonstrated by segmenting synthetic targets and actual kidney tumors from CT images acquired from Tartu University Hospital in Estonia. The findings indicate that COIN greatly surpasses established attribution methods, such as RISE, ScoreCAM, and LayerCAM, as well as an alternative counterfactual explanation method introduced by Singla et al. This evidence suggests that COIN is a promising approach for semantic segmentation of tumors in CT images, and presents a step forward in making deep learning applications more accessible and effective in healthcare, where annotated data is scarce., Comment: This work has been accepted to be presented to The 2nd World Conference on eXplainable Artificial Intelligence (xAI 2024), July 17-19, 2024 - Valletta, Malta

Health acoustic sounds such as coughs and breaths are known to contain useful health signals with significant potential for monitoring health and disease, yet are underexplored in the medical machine learning community. The existing deep learning systems for health acoustics are often narrowly trained and evaluated on a single task, which is limited by data and may hinder generalization to other tasks. To mitigate these gaps, we develop HeAR, a scalable self-supervised learning-based deep learning system using masked autoencoders trained on a large dataset of 313 million two-second long audio clips. Through linear probes, we establish HeAR as a state-of-the-art health audio embedding model on a benchmark of 33 health acoustic tasks across 6 datasets. By introducing this work, we hope to enable and accelerate further health acoustics research., Comment: 4 tables, 4 figures, 6 supplementary tables, 3 supplementary figures

Two-photon absorption in indirect gap semiconductors is an frequently encountered, but not well-understood phenomenon. To address this, the Real Density Matrix Approach is applied to describe two-photon absorption in silicon through the excitonic response to the interacting fields. This approach produces an analytical expression for the dispersion of the two-photon absorption coefficient for indirect-gap materials, and can be used to explain trends in reported experimental data for bulk silicon both old and new with minimal fitting.

The nature of enhanced photoemission in disordered and amorphous solids is an intriguing open question. A point in case is light emission in silicon, which occurs when the material is porous or nanostructured, but the effect is absent in the bulk crystalline phase, a phenomenon that is still not fully understood. In this work, we study structural photoemission in a heterogeneous cross-linked silicon glass, a material that represents an intermediate state between the amorphous and crystalline phases, characterized by a narrow distribution of structure sizes. This model system shows a clear dependence of photoemission on size and disorder across a broad range of energies. While phonon-assisted indirect optical transitions are insufficient to describe observable emissions, our experiments suggest these can be understood through electronic Raman scattering instead. This phenomenon, not commonly observed in crystalline semiconductors, is driven by structural disorder. We attribute photoemission in this disordered system to the presence of an excess electron density of states within the forbidden gap (Urbach bridge), where electrons occupy trapped states. Transitions from gap states to the conduction band are facilitated through electron-photon momentum matching, which resembles Compton scattering, but observed for visible light and driven by the enhanced momentum of a photon confined within the nanostructured domains. We interpret the light emission in structured silicon glass as resulting from electronic Raman scattering. These findings emphasize the role of photon momentum in the optical response of solids that display disorder at the nanoscale.