Search

Conformal Prediction is a robust framework that ensures reliable coverage across machine learning tasks. Although recent studies have applied conformal prediction to graph neural networks, they have largely emphasized post-hoc prediction set generation. Improving conformal prediction during the training stage remains unaddressed. In this work, we tackle this challenge from a denoising perspective by introducing SparGCP, which incorporates graph sparsification and a conformal prediction-specific objective into GNN training. SparGCP employs a parameterized graph sparsification module to filter out task-irrelevant edges, thereby improving conformal prediction efficiency. Extensive experiments on real-world graph datasets demonstrate that SparGCP outperforms existing methods, reducing prediction set sizes by an average of 32\% and scaling seamlessly to large networks on commodity GPUs.

We introduce the Byte Latent Transformer (BLT), a new byte-level LLM architecture that, for the first time, matches tokenization-based LLM performance at scale with significant improvements in inference efficiency and robustness. BLT encodes bytes into dynamically sized patches, which serve as the primary units of computation. Patches are segmented based on the entropy of the next byte, allocating more compute and model capacity where increased data complexity demands it. We present the first FLOP controlled scaling study of byte-level models up to 8B parameters and 4T training bytes. Our results demonstrate the feasibility of scaling models trained on raw bytes without a fixed vocabulary. Both training and inference efficiency improve due to dynamically selecting long patches when data is predictable, along with qualitative improvements on reasoning and long tail generalization. Overall, for fixed inference costs, BLT shows significantly better scaling than tokenization-based models, by simultaneously growing both patch and model size.

Neural Machine Translation (NMT) models are typically trained on datasets with limited exposure to Scientific, Technical and Educational domains. Translation models thus, in general, struggle with tasks that involve scientific understanding or technical jargon. Their performance is found to be even worse for low-resource Indian languages. Finding a translation dataset that tends to these domains in particular, poses a difficult challenge. In this paper, we address this by creating a multilingual parallel corpus containing more than 2.8 million rows of English-to-Indic and Indic-to-Indic high-quality translation pairs across 8 Indian languages. We achieve this by bitext mining human-translated transcriptions of NPTEL video lectures. We also finetune and evaluate NMT models using this corpus and surpass all other publicly available models at in-domain tasks. We also demonstrate the potential for generalizing to out-of-domain translation tasks by improving the baseline by over 2 BLEU on average for these Indian languages on the Flores+ benchmark. We are pleased to release our model and dataset via this link: https://huggingface.co/SPRINGLab.

Novel-view synthesis techniques achieve impressive results for static scenes but struggle when faced with the inconsistencies inherent to casual capture settings: varying illumination, scene motion, and other unintended effects that are difficult to model explicitly. We present an approach for leveraging generative video models to simulate the inconsistencies in the world that can occur during capture. We use this process, along with existing multi-view datasets, to create synthetic data for training a multi-view harmonization network that is able to reconcile inconsistent observations into a consistent 3D scene. We demonstrate that our world-simulation strategy significantly outperforms traditional augmentation methods in handling real-world scene variations, thereby enabling highly accurate static 3D reconstructions in the presence of a variety of challenging inconsistencies. Project page: https://alextrevithick.github.io/simvs, Comment: Project page: https://alextrevithick.github.io/simvs

We demonstrate a machine learning (ML) approach that accurately predicts the current-voltage behavior of 3D/2D-structured (FAMA)Pb(IBr)3/OABr hybrid organic-inorganic halide perovskite (HOIP) solar cells under AM1.5 illumination. Our neural network algorithm is trained on measured responses from several hundred HOIP solar cells, using three simple optical measurements of constituent HOIP films as input: optical transmission spectrum, spectrally-resolved photoluminescence, and time-resolved photoluminescence, from which we predict the open-circuit voltage (Voc), short-circuit current (Jsc), and fill factors (FF) values of solar cells that contain the HOIP active layers. Determined average prediction accuracies for 95 % of the predicted Voc, Jsc, and FF values are 91%, 94% and 89%, respectively, with R2 coefficients of determination of 0.47, 0.77, and 0.58, respectively. Quantifying the connection between ML predictions and physical parameters extracted from the measured HOIP films optical properties, allows us to identify the most significant parameters influencing the prediction results. With separate ML-classifying algorithms, we identify degraded solar cells using the same optical input data, achieving over 90% classification accuracy through support vector machine, cross entropy loss, and artificial neural network algorithms. To our knowledge, the demonstrated regression and classification work is the first to use ML to predict device photovoltaic properties solely from the optical properties of constituent materials., Comment: 36 pages, 6 figures

We propose a novel method for measuring the discrepancy between a set of samples and a desired posterior distribution for Bayesian inference. Classical methods for assessing sample quality like the effective sample size are not appropriate for scalable Bayesian sampling algorithms, such as stochastic gradient Langevin dynamics, that are asymptotically biased. Instead, the gold standard is to use the kernel Stein Discrepancy (KSD), which is itself not scalable given its quadratic cost in the number of samples. The KSD and its faster extensions also typically suffer from the curse-of-dimensionality and can require extensive tuning. To address these limitations, we develop the polynomial Stein discrepancy (PSD) and an associated goodness-of-fit test. While the new test is not fully convergence-determining, we prove that it detects differences in the first r moments in the Bernstein-von Mises limit. We empirically show that the test has higher power than its competitors in several examples, and at a lower computational cost. Finally, we demonstrate that the PSD can assist practitioners to select hyper-parameters of Bayesian sampling algorithms more efficiently than competitors., Comment: 17 Pages, 14 Figs

The evolution of multicellularity involved the transformation of a simple cell wall of unicellular ancestors into a complex, multifunctional extracellular matrix (ECM). A suitable model organism to study the formation and expansion of an ECM during ontogenesis is the multicellular green alga $Volvox~carteri$, which, along with the related volvocine algae, produces a complex, self-organized ECM composed of multiple substructures. These self-assembled ECMs primarily consist of hydroxyproline-rich glycoproteins, a major component of which is pherophorins. To investigate the geometry of the growing ECM, we fused the $yfp$ gene with the gene for pherophorin II (PhII) in $V.~carteri$. Confocal microscopy reveals PhII:YFP localization at key structures within the ECM, including the boundaries of compartments surrounding each somatic cell and the outer surface of the organism. Image analysis during the life cycle allows the stochastic geometry of those growing compartments to be quantified. We find that their areas and aspect ratios exhibit robust gamma distributions and exhibit a transition from a tight polygonal to a looser acircular packing geometry with stable eccentricity over time, evoking parallels and distinctions with the behavior of hydrated foams. These results provide a quantitative benchmark for addressing a general, open question in biology: How do cells produce structures external to themselves in a robust and accurate manner?, Comment: 14 pages, 12 figures + SI Appendix (21 pages, 11 figures)

Computing high dimensional potential surfaces for molecular and materials systems is considered to be a great challenge in computational chemistry with potential impact in a range of areas including fundamental prediction of reaction rates. In this paper we design and discuss an algorithm that has similarities to large language models in generative AI and natural language processing. Specifically, we represent a molecular system as a graph which contains a set of nodes, edges, faces etc. Interactions between these sets, which represent molecular subsystems in our case, are used to construct the potential energy surface for a reasonably sized chemical system with 51 dimensions. Essentially a family of neural networks that pertain to the graph-based subsystems, get the job done for this 51 dimensional system. We then ask if this same family of lower-dimensional neural networks can be transformed to provide accurate predictions for a 186 dimensional potential surface. We find that our algorithm does provide reasonably accurate results for this larger dimensional problem with sub-kcal/mol accuracy for the higher dimensional potential surface problem.

The accurate computational study of wavepacket nuclear dynamics is considered to be a classically intractable problem, particularly with increasing dimensions. Here we present two algorithms that, in conjunction with other methods developed by us, will form the basis for performing quantum nuclear dynamics in arbitrary dimensions. For one algorithm, we present a direct map between the Born-Oppenheimer Hamiltonian describing the wavepacket time-evolution and the control parameters of a spin-lattice Hamiltonian that describes the dynamics of qubit states in an ion-trap quantum computer. This map is exact for three qubits, and when implemented, the dynamics of the spin states emulate those of the nuclear wavepacket. However, this map becomes approximate as the number of qubits grow. In a second algorithm we present a general quantum circuit decomposition formalism for such problems using a method called the Quantum Shannon Decomposition. This algorithm is more robust and is exact for any number of qubits, at the cost of increased circuit complexity. The resultant circuit is implemented on IBM's quantum simulator (QASM) for 3-7 qubits. In both cases the wavepacket dynamics is found to be in good agreement with the classical result and the corresponding vibrational frequencies obtained from the wavepacket density time-evolution, are in agreement to within a few tenths of a wavenumbers.

Body-coupled powering (BCP) is an innovative wireless power transfer (WPT) technique, recently explored for its potential to deliver power to cutting-edge biomedical implants such as nerve and muscle stimulators. This paper demonstrates the efficient technique of designing WPT systems embedding BCP via galvanic coupling (G-BCP). The G-BCP configuration utilizes two metal circular rings surrounding the body area of interest as the transmitter (TX) electrodes required for galvanic (differential) excitation and a wireless implant as the receiver (RX) equipped with two electrodes for differential power reception accordingly. By focusing on the unique advantages of this approach - such as enhanced targeting accuracy, improved power transfer efficiency (PTE), and favorable tissue penetration characteristics, G-BCP emerges as a superior alternative to traditional WPT methods. A comprehensive analysis is conducted to obtain the optimized device parameters while simultaneously allowing flexible placement of implants at different depths and alignments. To substantiate the proposed design concept, a prototype was simulated in Ansys HFSS, employing a multi-layered tissue medium of 10mm radius and targeting the sciatic nerve of a rat. Impressively, this prototype achieves > 20% PTE at 1.25 GHz, with the implant (radius of RX electrodes = 1 mm) located 2 mm deep inside the tissue model having complex load impedance of Rload = 1000 Ohm and Cload = 5pF. Therefore, the G-BCP-based wirelessly powered microdevices are envisaged to be a key enabler in neural recording and stimulation, specifically for the peripheral nervous system, enhancing therapeutic outcomes and patient experiences., Comment: 5 pages

This paper analyzes the performance of Small Language Models (SLMs) and Vision Language Models (VLMs) and evaluates the trade-off between model performance and carbon emissions across 4 essential tasks: Image Captioning, Visual Question Answering (VQA), Dialogue Summarization and Text-to-SQL conversion. Various SLMs and VLMs belonging to the Qwen and LLaMA architecture family are chosen and variants based on model size in terms of the number of parameters, quantization level and fine-tuning parameters are evaluated. The model variant's performance and carbon emissions are calculated. To quantify the trade-off between model performance and carbon emissions, we introduce a novel metric called CEGI (Carbon Efficient Gain Index). This metric represents the carbon emission per unit percentage gain per million trainable parameters . This metric provides a normalized measure to compare model's efficiency in terms of performance improvement relative to their environmental cost. The experiment's outcome demonstrates that fine-tuning SLMs and VLMs can achieve performance levels comparable to Large Language Models (LLMs) while producing significantly less carbon emissions. Our findings suggest that the marginal gains in accuracy from larger models do not justify the substantial increase in carbon emissions. Leveraging lower-bit quantization levels, the proposed metric further enhances energy efficiency without compromising performance. This study highlights balancing high performance and environmental sustainability. It offers a valuable metric for selecting models suitable for environmentally-friendly AI development.

The Ceresa cycle is a canonical algebraic $1$-cycle on the Jacobian of an algebraic curve. We construct an algorithm which, given a curve over a number field, often provides a certificate that the Ceresa cycle is non-torsion, without relying on the presence of any additional symmetries of the curve. Under the hypothesis that the Sato--Tate group is the whole of $\operatorname*{GSp}$, we prove that if the Ceresa class (the image of the Ceresa cycle in \'{e}tale cohomology) is non-torsion, then the algorithm will eventually terminate with a certificate attesting to this fact.

This paper examines the impact of racial discrimination in hiring on employment, wage, and wealth disparities between black and white workers. Using a labor search-and-matching model with racially prejudiced and non-prejudiced firms, we show that labor market frictions sustain discriminatory practices as an equilibrium outcome. These practices account for 44% to 52% of the average wage gap and 16% of the median wealth gap. Discriminatory hiring also amplifies unemployment and wage volatility for black workers, increasing their labor market risks over business cycles. Eliminating prejudiced firms reduces these disparities and improves black workers' welfare, though it slightly decreases overall economic welfare.

As machine learning gets deployed more and more widely, and model sizes continue to grow, improving computational efficiency during model inference has become a key challenge. In many commonly used model architectures, including Transformers, a significant portion of the inference computation is comprised of exponential non-linearities such as Softmax. In this work, we develop QuAKE, a collection of novel operators that leverage certain properties of IEEE-754 floating point representations to quickly approximate the exponential function without requiring specialized hardware, extra memory, or precomputation. We propose optimizations that enhance the efficiency of QuAKE in commonly used exponential non-linearities such as Softmax, GELU, and the Logistic function. Our benchmarks demonstrate substantial inference speed improvements between 10% and 35% on server CPUs, and 5% and 45% on embedded and mobile-scale CPUs for a variety of model architectures and sizes. Evaluations of model performance on standard datasets and tasks from various domains show that QuAKE operators are able to provide sizable speed benefits with little to no loss of performance on downstream tasks.

We propose a novel ansatz, where the full black hole geometry is written as a linear in mass perturbation of the associated extremal black hole base. Contrary to its "standard" version, the corresponding "extremal Kerr-Schild form" is no longer restricted to special algebraic type spacetimes, and is applicable to numerous black hole solutions with matter, such as the charged Kerr-NUT-(A)dS spacetimes, black holes of $D=5$ minimal gauged supergravity, or the charged dilaton-axion rotating solutions. This ansatz is likely to find its applications in black hole perturbation theory, shed new light on the CFT description of non-extremal black holes, as well as be useful for constructing new exact solutions., Comment: 9 pages, no figures

We propose a diffusion model-based approach, FloAtControlNet to generate cinemagraphs composed of animations of human clothing. We focus on human clothing like dresses, skirts and pants. The input to our model is a text prompt depicting the type of clothing and the texture of clothing like leopard, striped, or plain, and a sequence of normal maps that capture the underlying animation that we desire in the output. The backbone of our method is a normal-map conditioned ControlNet which is operated in a training-free regime. The key observation is that the underlying animation is embedded in the flow of the normal maps. We utilize the flow thus obtained to manipulate the self-attention maps of appropriate layers. Specifically, the self-attention maps of a particular layer and frame are recomputed as a linear combination of itself and the self-attention maps of the same layer and the previous frame, warped by the flow on the normal maps of the two frames. We show that manipulating the self-attention maps greatly enhances the quality of the clothing animation, making it look more natural as well as suppressing the background artifacts. Through extensive experiments, we show that the method proposed beats all baselines both qualitatively in terms of visual results and user study. Specifically, our method is able to alleviate the background flickering that exists in other diffusion model-based baselines that we consider. In addition, we show that our method beats all baselines in terms of RMSE and PSNR computed using the input normal map sequences and the normal map sequences obtained from the output RGB frames. Further, we show that well-established evaluation metrics like LPIPS, SSIM, and CLIP scores that are generally for visual quality are not necessarily suitable for capturing the subtle motions in human clothing animations.

Graphic designs are an effective medium for visual communication. They range from greeting cards to corporate flyers and beyond. Off-late, machine learning techniques are able to generate such designs, which accelerates the rate of content production. An automated way of evaluating their quality becomes critical. Towards this end, we introduce Design-o-meter, a data-driven methodology to quantify the goodness of graphic designs. Further, our approach can suggest modifications to these designs to improve its visual appeal. To the best of our knowledge, Design-o-meter is the first approach that scores and refines designs in a unified framework despite the inherent subjectivity and ambiguity of the setting. Our exhaustive quantitative and qualitative analysis of our approach against baselines adapted for the task (including recent Multimodal LLM-based approaches) brings out the efficacy of our methodology. We hope our work will usher more interest in this important and pragmatic problem setting., Comment: Accepted to WACV 2025. Project page: https://sahilg06.github.io/Design-o-meter/

Understanding the performance of electrochemical energy storage systems requires probing the electrochemical properties at each layer and interface during cell operation. While traditional onboard and operando methods can measure impedance, voltage, or capacity, they lack spatial resolution to pinpoint the properties to specific layers and interfaces. In this work, we describe an approach of using thermal waves to measure entropy change, transport resistance, and charge-transfer resistance with depth resolution of a few microns within an electrochemical cell. We achieve this by relating heat generation at multiple harmonics of an AC current to electrochemical processes and leveraging frequency dependence of thermal penetration depth for spatial resolution. We name this frequency domain spectroscopy of the thermal signatures of the electrochemical processes measured at multiple harmonics of the alternating current as Multi-harmonic ElectroThermal Spectroscopy (METS). This technique enables isolation and measurement of solvation entropy at individual electrode-electrolyte interfaces from the first harmonic (1{\omega}) thermal signature and resolution of the overall interfacial impedance into charge-transfer and interface transport resistance components from the second harmonic (2{\omega}) thermal signature. From this, we also demonstrate an operando measurement of the growth of the solid-electrolyte interphase (SEI) layer at the lithium-electrolyte interface and show that two chemically similar electrodes can have significantly different interfacial transport resistance based on the preparation of the electrodes. Additionally, the method is not specific to lithium-ion chemistry and can therefore be generalized for all electrochemical systems of interest.

We consider the problem of conditional text-to-image synthesis with diffusion models. Most recent works need to either finetune specific parts of the base diffusion model or introduce new trainable parameters, leading to deployment inflexibility due to the need for training. To address this gap in the current literature, we propose our method called TINTIN: Test-time Conditional Text-to-Image Synthesis using Diffusion Models which is a new training-free test-time only algorithm to condition text-to-image diffusion model outputs on conditioning factors such as color palettes and edge maps. In particular, we propose to interpret noise predictions during denoising as gradients of an energy-based model, leading to a flexible approach to manipulate the noise by matching predictions inferred from them to the ground truth conditioning input. This results in, to the best of our knowledge, the first approach to control model outputs with input color palettes, which we realize using a novel color distribution matching loss. We also show this test-time noise manipulation can be easily extensible to other types of conditioning, e.g., edge maps. We conduct extensive experiments using a variety of text prompts, color palettes, and edge maps and demonstrate significant improvement over the current state-of-the-art, both qualitatively and quantitatively.

Artificial Intelligence models encoding biology and chemistry are opening new routes to high-throughput and high-quality in-silico drug development. However, their training increasingly relies on computational scale, with recent protein language models (pLM) training on hundreds of graphical processing units (GPUs). We introduce the BioNeMo Framework to facilitate the training of computational biology and chemistry AI models across hundreds of GPUs. Its modular design allows the integration of individual components, such as data loaders, into existing workflows and is open to community contributions. We detail technical features of the BioNeMo Framework through use cases such as pLM pre-training and fine-tuning. On 256 NVIDIA A100s, BioNeMo Framework trains a three billion parameter BERT-based pLM on over one trillion tokens in 4.2 days. The BioNeMo Framework is open-source and free for everyone to use.

An oracle determines whether the output of a program for executed test cases is correct. For machine learning programs, such an oracle is often unavailable or impractical to apply. Metamorphic testing addresses this by using metamorphic relations (MRs), which are essential properties of the software under test, to verify program correctness. Prioritizing MRs enhances fault detection effectiveness and improves testing efficiency. However, fault-based MR prioritization is costly, and code coverage-based approaches often yield inaccurate results. To address these challenges, we propose a statement centrality-based prioritization approach that leverages diversity in the execution profiles of source and follow-up test cases. Our approach improves fault detection effectiveness by up to 31% compared to code coverage-based methods and reduces fault detection time by 29% compared to random MR execution. Overall, it increases the average fault detection rate, reduces fault detection time, and enhances testing efficiency., Comment: arXiv admin note: text overlap with arXiv:2109.09798, arXiv:2209.00162

The Fermi-Hubbard model (FHM) is a simple yet rich model of strongly interacting electrons with complex dynamics and a variety of emerging quantum phases. These properties make it a compelling target for digital quantum simulation. Trotterization-based quantum simulations have shown promise, but implementations on current hardware are limited by noise, necessitating error mitigation techniques like circuit optimization and post-selection. A mapping of the FHM to a Z2 LGT was recently proposed that restricts the dynamics to a subspace protected by additional symmetries, and its ability for post-selection error mitigation was verified through noisy classical simulations. In this work, we propose and demonstrate a suite of algorithm-hardware co-design strategies on a trapped-ion quantum computer, targeting two key aspects of NISQ-era quantum simulation: circuit compilation and error mitigation. In particular, a novel combination of iteratively preconditioned gradient descent (IPG) and subsystem von Neumann Entropy compression reduces the 2-qubit gate count of FHM quantum simulation by 35%, consequently doubling the number of simulatable Trotter steps when used in tandem with error mitigation based on conserved symmetries, debiasing and sharpening techniques. Our work demonstrates the value of algorithm-hardware co-design to operate digital quantum simulators at the threshold of maximum circuit depths allowed by current hardware, and is broadly generalizable to strongly correlated systems in quantum chemistry and materials science.

In this work, we show that the class of multivariate degree-$d$ polynomials mapping $\{0,1\}^{n}$ to any Abelian group $G$ is locally correctable with $\widetilde{O}_{d}((\log n)^{d})$ queries for up to a fraction of errors approaching half the minimum distance of the underlying code. In particular, this result holds even for polynomials over the reals or the rationals, special cases that were previously not known. Further, we show that they are locally list correctable up to a fraction of errors approaching the minimum distance of the code. These results build on and extend the prior work of the authors [ABPSS24] (STOC 2024) who considered the case of linear polynomials and gave analogous results. Low-degree polynomials over the Boolean cube $\{0,1\}^{n}$ arise naturally in Boolean circuit complexity and learning theory, and our work furthers the study of their coding-theoretic properties. Extending the results of [ABPSS24] from linear to higher-degree polynomials involves several new challenges and handling them gives us further insights into properties of low-degree polynomials over the Boolean cube. For local correction, we construct a set of points in the Boolean cube that lie between two exponentially close parallel hyperplanes and is moreover an interpolating set for degree-$d$ polynomials. To show that the class of degree-$d$ polynomials is list decodable up to the minimum distance, we stitch together results on anti-concentration of low-degree polynomials, the Sunflower lemma, and the Footprint bound for counting common zeroes of polynomials. Analyzing the local list corrector of [ABPSS24] for higher degree polynomials involves understanding random restrictions of non-zero degree-$d$ polynomials on a Hamming slice. In particular, we show that a simple random restriction process for reducing the dimension of the Boolean cube is a suitably good sampler for Hamming slices., Comment: 64 pages, To appear in SODA 2025, deleted image files

The development of large language models (LLMs) has expanded to multi-modal systems capable of processing text, images, and speech within a unified framework. Training these models demands significantly larger datasets and computational resources compared to text-only LLMs. To address the scaling challenges, we introduce Mixture-of-Transformers (MoT), a sparse multi-modal transformer architecture that significantly reduces pretraining computational costs. MoT decouples non-embedding parameters of the model by modality -- including feed-forward networks, attention matrices, and layer normalization -- enabling modality-specific processing with global self-attention over the full input sequence. We evaluate MoT across multiple settings and model scales. In the Chameleon 7B setting (autoregressive text-and-image generation), MoT matches the dense baseline's performance using only 55.8\% of the FLOPs. When extended to include speech, MoT reaches speech performance comparable to the dense baseline with only 37.2\% of the FLOPs. In the Transfusion setting, where text and image are trained with different objectives, a 7B MoT model matches the image modality performance of the dense baseline with one third of the FLOPs, and a 760M MoT model outperforms a 1.4B dense baseline across key image generation metrics. System profiling further highlights MoT's practical benefits, achieving dense baseline image quality in 47.2\% of the wall-clock time and text quality in 75.6\% of the wall-clock time (measured on AWS p4de.24xlarge instances with NVIDIA A100 GPUs).

The widespread availability of maps has enabled the development of numerous location-based applications, including navigation, ride-sharing, fitness tracking, gaming, robotics, and augmented reality. Today, the maps that power these services are predominantly controlled by a few large corporations and mostly cover outdoor spaces. As the use of these applications expands and indoor localization technologies advance, we are seeing the need for a scalable, federated location management system that can extend into private spaces. We introduce OpenFLAME (Open Federated Localization and Mapping Engine), the first federated and decentralized localization service. OpenFLAME links servers that handle localization for specific regions, providing applications with a seamless global view. Creating a federated localization system poses challenges, such as discovering the appropriate servers for a region and integrating services managed by independent providers. To address these issues and ensure scalability, we leverage Domain Name System (DNS) for service discovery and implement map abstractions to retrieve and merge locations across different maps. Our trace-driven study demonstrates that federated localization across remote servers is feasible with acceptable query latencies. To highlight the potential of the system, we developed an augmented reality navigation application for a large indoor space, showing that OpenFLAME can successfully power location-based applications.

We develop a robust translation model for four low-resource Indic languages: Khasi, Mizo, Manipuri, and Assamese. Our approach includes a comprehensive pipeline from data collection and preprocessing to training and evaluation, leveraging data from WMT task datasets, BPCC, PMIndia, and OpenLanguageData. To address the scarcity of bilingual data, we use back-translation techniques on monolingual datasets for Mizo and Khasi, significantly expanding our training corpus. We fine-tune the pre-trained NLLB 3.3B model for Assamese, Mizo, and Manipuri, achieving improved performance over the baseline. For Khasi, which is not supported by the NLLB model, we introduce special tokens and train the model on our Khasi corpus. Our training involves masked language modelling, followed by fine-tuning for English-to-Indic and Indic-to-English translations., Comment: Published in WMT 2024. Low-Resource Indic Language Translation Shared Task

We consider the problems of testing and learning an unknown $n$-qubit Hamiltonian $H$ from queries to its evolution operator $e^{-iHt}$ under the normalized Frobenius norm. We prove: 1. Local Hamiltonians: We give a tolerant testing protocol to decide if $H$ is $\epsilon_1$-close to $k$-local or $\epsilon_2$-far from $k$-local, with $O(1/(\epsilon_2-\epsilon_1)^{4})$ queries, solving open questions posed in a recent work by Bluhm et al. For learning a $k$-local $H$ up to error $\epsilon$, we give a protocol with query complexity $\exp(O(k^2+k\log(1/\epsilon)))$ independent of $n$, by leveraging the non-commutative Bohnenblust-Hille inequality. 2. Sparse Hamiltonians: We give a protocol to test if $H$ is $\epsilon_1$-close to being $s$-sparse (in the Pauli basis) or $\epsilon_2$-far from being $s$-sparse, with $O(s^{6}/(\epsilon_2^2-\epsilon_1^2)^{6})$ queries. For learning up to error $\epsilon$, we show that $O(s^{4}/\epsilon^{8})$ queries suffice. 3. Learning without memory: The learning results stated above have no dependence on $n$, but require $n$-qubit quantum memory. We give subroutines that allow us to learn without memory; increasing the query complexity by a $(\log n)$-factor in the local case and an $n$-factor in the sparse case. 4. Testing without memory: We give a new subroutine called Pauli hashing, which allows one to tolerantly test $s$-sparse Hamiltonians with $O(s^{14}/(\epsilon_2^2-\epsilon_1^2)^{18})$ queries. A key ingredient is showing that $s$-sparse Pauli channels can be tolerantly tested under the diamond norm with $O(s^2/(\epsilon_2-\epsilon_1)^6)$ queries. Along the way, we prove new structural theorems for local and sparse Hamiltonians. We complement our learning results with polynomially weaker lower bounds. Furthermore, our algorithms use short time evolutions and do not assume prior knowledge of the terms in the support of the Pauli spectrum of $H$., Comment: 45 pages. This work subsumes a prior work by the third author (arXiv:2404.06282)

Eye tracking technology offers great potential for improving road safety. It is already being built into vehicles, namely cars and trucks. When this technology is integrated into transit service vehicles, employees, i.e., bus drivers, will be subject to being eye tracked on their job. Although there is much research effort advancing algorithms for eye tracking in transportation, less is known about how end users perceive this technology, especially when interacting with it in an employer-mandated context. In this first study of its kind, we investigated transit bus operators' perceptions of eye tracking technology. From a methodological perspective, we introduce a mixed methods approach where participants experience the technology first-hand and then reflect on their experience while viewing a playback of the recorded data. Thematic analysis of the interview transcripts reveals interesting potential uses of eye tracking in this work context and surfaces transit operators' fears and concerns about this technology.

This work explores conditions under which multi-finger grasping algorithms can attain robust sim-to-real transfer. While numerous large datasets facilitate learning generative models for multi-finger grasping at scale, reliable real-world dexterous grasping remains challenging, with most methods degrading when deployed on hardware. An alternate strategy is to use discriminative grasp evaluation models for grasp selection and refinement, conditioned on real-world sensor measurements. This paradigm has produced state-of-the-art results for vision-based parallel-jaw grasping, but remains unproven in the multi-finger setting. In this work, we find that existing datasets and methods have been insufficient for training discriminitive models for multi-finger grasping. To train grasp evaluators at scale, datasets must provide on the order of millions of grasps, including both positive and negative examples, with corresponding visual data resembling measurements at inference time. To that end, we release a new, open-source dataset of 3.5M grasps on 4.3K objects annotated with RGB images, point clouds, and trained NeRFs. Leveraging this dataset, we train vision-based grasp evaluators that outperform both analytic and generative modeling-based baselines on extensive simulated and real-world trials across a diverse range of objects. We show via numerous ablations that the key factor for performance is indeed the evaluator, and that its quality degrades as the dataset shrinks, demonstrating the importance of our new dataset. Project website at: https://sites.google.com/view/get-a-grip-dataset.