Search

In this paper, we propose the first method to modify the prosodic features of a given speech signal using actor-critic reinforcement learning strategy. Our approach uses a Bayesian framework to identify contiguous segments of importance that links segments of the given utterances to perception of emotions in humans. We train a neural network to produce the variational posterior of a collection of Bernoulli random variables; our model applies a Markov prior on it to ensure continuity. A sample from this distribution is used for downstream emotion prediction. Further, we train the neural network to predict a soft assignment over emotion categories as the target variable. In the next step, we modify the prosodic features (pitch, intensity, and rhythm) of the masked segment to increase the score of target emotion. We employ an actor-critic reinforcement learning to train the prosody modifier by discretizing the space of modifications. Further, it provides a simple solution to the problem of gradient computation through WSOLA operation for rhythm manipulation. Our experiments demonstrate that this framework changes the perceived emotion of a given speech utterance to the target. Further, we show that our unified technique is on par with state-of-the-art emotion conversion models from supervised and unsupervised domains that require pairwise training., Comment: 7 pages, 10 figures

We study push-pull rumour spreading in small-world models for social networks where the degrees follow a power-law. In a non-geometric setting Fountoulakis, Panagiotou and Sauerwald have shown that rumours always spread fast (SODA 2012). On the other hand, Janssen and Mehrabian have found that rumours spread slowly in a spatial preferential attachment model (SIDMA 2017). We study the question systematically for the model of geometric inhomogeneous random graphs (GIRGs), which has been found to be a good theoretical and empirical fit for social networks. Our result is two-fold: with classical Euclidean geometry both slow and fast rumour spreading may occur, depending on the exponent of the power law and the prevalence of weak ties in the networks, and we fully characterise the phase boundaries between those two regimes. Depending on the parameters, fast spreading may either mean polylogarithmic time or even doubly logarithmic time. Secondly, we show that rumour spreading is always fast in a non-metric geometry. The considered non-metric geometry allows to model social connections where resemblance of vertices in a single attribute, such as familial kinship, already strongly indicates the presence of an edge. Classical Euclidean Geometry fails to capture such ties. For some regimes in the Euclidean setting, the efficient pathways for spreading rumours differ from previously identified paths. A vertex of degree $d$ can transmit the rumour efficiently to a vertex of larger degree by a chain of length $3$, where one of the two intermediaries has constant degree, and the other has degree $d^{c}$ for some constant $c<1$., Comment: 40 pages

Vertical Take-Off and Landing (VTOL) vehicles have gained immense popularity in the delivery drone market and are now being developed for passenger transportation in urban areas to efficiently enable Urban Air Mobility (UAM). UAM aims to utilize the urban airspace \hidetxt{vertical dimension} to address the problem of heavy road congestion in dense urban cities. VTOL vehicles require vertiport terminals for landing, take-off, passengers boarding or deboarding, refuelling (or charging), and maintenance. An efficient scheduling algorithm is essential to maximize the throughput of the vertiport terminal (vertiminal)\hidetxt{ as well as efficient use of airspace} while maintaining safety protocols to handle the UAM traffic. While traditional departure and taxiing operations can be applied in the context of vertiminal, specific algorithms are required for take-off and landing schedules. Unlike fixed-wing aircraft that require a runway to take-off and climb in a single direction, VTOL vehicles can approach and climb in several directions. We propose a Mixed Integer Linear Program (MILP) formulation to schedule flights for taxiing, climbing (or approaching) using multiple directions after take-off (before landing) and turnaround on gates. We also derived equations to thoroughly analyze the throughput capacity of a vertiminal considering all its core elements. We have shown that our MILP can achieve the maximum throughput obtained through the equations. Given the input parameters, our analysis can be used to analyze the capacity of a vertiminal without running any simulation, while our MILP can be used to get the most efficient schedule., Comment: Extension of conference work "Integrated Taxiing and TLOF pad Scheduling Using Different Surface Directions with Fairness Analysis" published in ITSC 2023. DOI: https://doi.org/10.1109/ITSC57777.2023.10422484

We present Segment Anything Model 2 (SAM 2), a foundation model towards solving promptable visual segmentation in images and videos. We build a data engine, which improves model and data via user interaction, to collect the largest video segmentation dataset to date. Our model is a simple transformer architecture with streaming memory for real-time video processing. SAM 2 trained on our data provides strong performance across a wide range of tasks. In video segmentation, we observe better accuracy, using 3x fewer interactions than prior approaches. In image segmentation, our model is more accurate and 6x faster than the Segment Anything Model (SAM). We believe that our data, model, and insights will serve as a significant milestone for video segmentation and related perception tasks. We are releasing our main model, dataset, as well as code for model training and our demo., Comment: Website: https://ai.meta.com/sam2

The location of the baryon acoustic oscillation (BAO) feature in the two-point correlation function (2PCF) of matter produces a standard ruler that is useful for the measurement of the expansion history of the Universe. Inspired by the possibility of reconstructing the positions of protohalos in the initial density field with a novel method rooted in optimal transport theory, we revisit the BAO signal in the protohalo correlation function. Our work examines the performance of a template 2PCF built on a tracer bias relation that includes scale dependence -- a term that can be motivated by peaks theory or a general bias expansion. Working in protohalos, halos, and the linear combination of the protohalo and matter fields that is motivated by the continuity equation, we demonstrate that this model accurately captures the shape of the BAO feature and improves the precision of the BAO scale measurement relative to a model that does not include scale-dependent bias by 47% in protohalos, 15% in halos, and 14% in the linear combination of the protohalo and matter fields. Allowing for scale dependence does not appear to introduce any shift in the BAO feature. The precision of the BAO distance scale estimate is highest with the linear combination of the protohalo and matter fields, which offers a factor of 3.5 improvement over Eulerian-space measurements and a factor of 4-8 improvement over the estimate made with protohalos alone., Comment: 13 pages, 7 figures, submitted to Physical Review D

Fixed-wing UAVs have transformed the transportation system with their high flight speed and long endurance, yet their safe operation in increasingly cluttered environments depends heavily on effective collision avoidance techniques. This paper presents a novel method for safely navigating an aircraft along a desired route while avoiding moving obstacles. We utilize a class of control barrier functions (CBFs) based on collision cones to ensure the relative velocity between the aircraft and the obstacle consistently avoids a cone of vectors that might lead to a collision. By demonstrating that the proposed constraint is a valid CBF for the aircraft, we can leverage its real-time implementation via Quadratic Programs (QPs), termed the CBF-QPs. Validation includes simulating control law along trajectories, showing effectiveness in both static and moving obstacle scenarios., Comment: 4 Pages, 3 figures. Presented at CyPhySS, 2024, Bangalore. arXiv admin note: text overlap with arXiv:2303.15871

We report a gravitational-wave parameter estimation algorithm, AMPLFI, based on likelihood-free inference using normalizing flows. The focus of AMPLFI is to perform real-time parameter estimation for candidates detected by machine-learning based compact binary coalescence search, Aframe. We present details of our algorithm and optimizations done related to data-loading and pre-processing on accelerated hardware. We train our model using binary black-hole (BBH) simulations on real LIGO-Virgo detector noise. Our model has $\sim 6$ million trainable parameters with training times $\lesssim 24$ hours. Based on online deployment on a mock data stream of LIGO-Virgo data, Aframe + AMPLFI is able to pick up BBH candidates and infer parameters for real-time alerts from data acquisition with a net latency of $\sim 6$s., Comment: Submitted to MLST

The clever hybridization of quantum computing concepts and evolutionary algorithms (EAs) resulted in a new field called quantum-inspired evolutionary algorithms (QIEAs). Unlike traditional EAs, QIEAs employ quantum bits to adopt a probabilistic representation of the state of a feature in a given solution. This unprecedented feature enables them to achieve better diversity and perform global search, effectively yielding a tradeoff between exploration and exploitation. We conducted a comprehensive survey across various publishers and gathered 56 papers. We thoroughly analyzed these publications, focusing on the novelty elements and types of heuristics employed by the extant quantum-inspired evolutionary algorithms (QIEAs) proposed to solve the feature subset selection (FSS) problem. Importantly, we provided a detailed analysis of the different types of objective functions and popular quantum gates, i.e., rotation gates, employed throughout the literature. Additionally, we suggested several open research problems to attract the attention of the researchers., Comment: 43 pages, 13 tables, 5 figures

In this work we study the problem of using machine-learned predictions to improve the performance of online algorithms. We consider two classical problems, ski rental and non-clairvoyant job scheduling, and obtain new online algorithms that use predictions to make their decisions. These algorithms are oblivious to the performance of the predictor, improve with better predictions, but do not degrade much if the predictions are poor., Comment: Conference version appeared in Neurips 2018

Quantum computing is moving swiftly from theoretical to practical applications, making it crucial to establish a significant quantum advantage. Despite substantial investments, access to quantum devices is still limited, with users facing issues like long wait times and inefficient resource management. Unlike the mature cloud solutions for classical computing, quantum computing lacks effective infrastructure for resource optimization. We propose a Quantum Resource Infrastructure Orchestrator (QRIO), a state-of-the-art cloud resource manager built on Kubernetes that is tailored to quantum computing. QRIO seeks to democratize access to quantum devices by providing customizable, user-friendly, open-source resource management. QRIO's design aims to ensure equitable access, optimize resource utilization, and support diverse applications, thereby speeding up innovation and making quantum computing more accessible and efficient to a broader user base. In this paper, we discuss QRIO's various features and evaluate its capability in several representative usecases., Comment: To appear at the IEEE International Symposium on Workload Characterization, 2024

The SNO+ detector operated initially as a water Cherenkov detector. The implementation of a sealed covergas system midway through water data taking resulted in a significant reduction in the activity of $^{222}$Rn daughters in the detector and allowed the lowest background to the solar electron scattering signal above 5 MeV achieved to date. This paper reports an updated SNO+ water phase $^8$B solar neutrino analysis with a total livetime of 282.4 days and an analysis threshold of 3.5 MeV. The $^8$B solar neutrino flux is found to be $\left(2.32^{+0.18}_{-0.17}\text{(stat.)}^{+0.07}_{-0.05}\text{(syst.)}\right)\times10^{6}$ cm$^{-2}$s$^{-1}$ assuming no neutrino oscillations, or $\left(5.36^{+0.41}_{-0.39}\text{(stat.)}^{+0.17}_{-0.16}\text{(syst.)} \right)\times10^{6}$ cm$^{-2}$s$^{-1}$ assuming standard neutrino oscillation parameters, in good agreement with both previous measurements and Standard Solar Model Calculations. The electron recoil spectrum is presented above 3.5 MeV.

We report on the second observing program of the active galactic nucleus NGC 4151 with simultaneous Imaging X-ray Polarimetry Explorer (IXPE; {\sim}750 ks), NuSTAR ({\sim}60 ks), XMM-Newton ({\sim}75 ks), and NICER ({\sim}65 ks) pointings. NGC 4151 is the first Type 1 radio-quiet Seyfert galaxy with constrained polarization properties for the X-ray corona. Despite the lower flux state in which the source is re-observed and the resulting higher contribution of the constant reflection component in the IXPE energy band, our results are in agreement with the first detection. From polarimetric analysis, a polarization degree {\Pi} = 4.7 {\pm} 1.3 percent and angle {\Psi} = 77{\deg} {\pm} 8{\deg} east of north (68 percent confidence level) are derived in the 2.0 - 8.0 keV energy range. Combining the two observations leads to polarization properties that are more constrained than those of the individual detections, showing {\Pi} = 4.5 {\pm} 0.9 percent and {\Psi} = 81{\deg} {\pm} 6{\deg} (with detection significance {\sim}4.6{\sigma}). The observed polarization angle aligns very well with the radio emission in this source, supporting, together with the significant polarization degree, a slab or wedge geometry for the X-ray corona. However, a switch in the polarization angle at low energies (37{\deg} {\pm} 7{\deg} in the 2 - 3.5 keV bin) suggests the presence of another component. When it is included in the spectro-polarimetric fit, a high polarization degree disfavors an interpretation in terms of a leakage through the absorbers, rather pointing to scattering from some kind of mirror., Comment: 9 pages, 8 figures, 3 tables. Accepted to Astronomy & Astrophysics

Molecular HIV Surveillance (MHS) has been described as key to enabling rapid responses to HIV outbreaks. It operates by linking individuals with genetically similar viral sequences, which forms a network. A major limitation of MHS is that it depends on sequence collection, which very rarely covers the entire population of interest. Ignoring missing data by conducting complete case analysis--which assumes that the observed network is complete--has been shown to result in significantly biased estimates of network properties. We use MHS to investigate disease dynamics of the HIV epidemic in Miami-Dade County (MDC) among men who have sex with men (MSM)--only 30.1% have a reported sequence. To do so, we present an approach for making Bayesian inferences on partially observed networks. Through a simulation study, we demonstrate a reduction in error of 43%-63% between our estimates and complete case analyses. We estimate increased mixing between MSM communities in MDC, defined by race and transmission risk compared to the results based on complete case analysis. Our approach makes use of a flexible network model--congruence class model--to overcome the high computational burden of previously reported Bayesian approaches to estimate network properties from partially observed networks., Comment: 19 pages; 6 figures; 2 tables

Skin diseases affect over a third of the global population, yet their impact is often underestimated. Automating skin disease classification to assist doctors with their prognosis might be difficult. Nevertheless, due to efficient feature extraction pipelines, deep learning techniques have shown much promise for various tasks, including dermatological disease identification. This study uses a skin disease dataset with 31 classes and compares it with all versions of Vision Transformers, Swin Transformers and DivoV2. The analysis is also extended to compare with benchmark convolution-based architecture presented in the literature. Transfer learning with ImageNet1k weights on the skin disease dataset contributes to a high test accuracy of 96.48\% and an F1-Score of 0.9727 using DinoV2, which is almost a 10\% improvement over this data's current benchmark results. The performance of DinoV2 was also compared for the HAM10000 and Dermnet datasets to test the model's robustness, and the trained model overcomes the benchmark results by a slight margin in test accuracy and in F1-Score on the 23 and 7 class datasets. The results are substantiated using explainable AI frameworks like GradCAM and SHAP, which provide precise image locations to map the disease, assisting dermatologists in early detection, prompt prognosis, and treatment., Comment: Submitted to Computers in Biology and Medicine

Rashba-Dresselhaus effects, which originate from spin-orbit coupling and allow for spin manipulations, are actively explored in materials following the pursuit of spintronics and quantum computing. However, materials that possess practically significant Rashba-Dresselhaus effects often contain toxic elements and offer little opportunity for tunability of the effects. We used first-principles simulations to reveal that the recently discovered halide ferroelectrics in the CsGeX$_3$ (X=Cl, Br, I) family possess large and tunable Rashba-Dresselhaus effects. In particular, they give origin to the spin splitting of up to 171meV in valence band of CsGeI$_3$. The value is chemically tunable and can decrease by 25% and 70% for CsGeBr$_3$ and CsGeCl$_3$, respectively. Such chemical tunability could result in engineering of desired values through solid solution technique. Application of electric field was found to result in structural changes that could both decrease and increase spin splitting leading to electrical tunability of the effect. In the vicinity of conduction and valence band extrema, the spin textures are mostly of Rashba type which is promising for spin-to-charge conversion applications. The spin directions are coupled with the polarization direction leading to Rashba-ferroelectricity co-functionality. Our work identifies lead-free perovskite halides as excellent candidates for spin-based applications and is likely to stimulate further research in this direction., Comment: 8 pages

Mechanistic interpretability aims to reverse engineer the computation performed by a neural network in terms of its internal components. Although there is a growing body of research on mechanistic interpretation of neural networks, the notion of a mechanistic interpretation itself is often ad-hoc. Inspired by the notion of abstract interpretation from the program analysis literature that aims to develop approximate semantics for programs, we give a set of axioms that formally characterize a mechanistic interpretation as a description that approximately captures the semantics of the neural network under analysis in a compositional manner. We use these axioms to guide the mechanistic interpretability analysis of a Transformer-based model trained to solve the well-known 2-SAT problem. We are able to reverse engineer the algorithm learned by the model -- the model first parses the input formulas and then evaluates their satisfiability via enumeration of different possible valuations of the Boolean input variables. We also present evidence to support that the mechanistic interpretation of the analyzed model indeed satisfies the stated axioms.

We present a recent calculation of the quark and gluon contributions to the angular momentum of a composite spin -$1/2$ state in QCD. The state we consider is a quark dressed with a gluon, and we use the two-component framework in light-front Hamiltonian QCD. We compare the results from different decompositions available in the literature. We also present the angular momentum distributions., Comment: Prepared for 31 st International Workshop on Deep Inelastic Scattering and Related Phenomena, Grenoble, France, April 8-12, 2024

As social media platforms grow, understanding the underlying reasons behind events and statements becomes crucial for businesses, policymakers, and researchers. This research explores the integration of Knowledge Graphs (KGs) with Large Language Models (LLMs) to perform causal analysis of tweets dataset. The LLM aided analysis techniques often lack depth in uncovering the causes driving observed effects. By leveraging KGs and LLMs, which encode rich semantic relationships and temporal information, this study aims to uncover the complex interplay of factors influencing causal dynamics and compare the results obtained using GPT-3.5 Turbo. We employ a Retrieval-Augmented Generation (RAG) model, utilizing a KG stored in a Neo4j (a.k.a PRAGyan) data format, to retrieve relevant context for causal reasoning. Our approach demonstrates that the KG-enhanced LLM RAG can provide improved results when compared to the baseline LLM (GPT-3.5 Turbo) model as the source corpus increases in size. Our qualitative analysis highlights the advantages of combining KGs with LLMs for improved interpretability and actionable insights, facilitating informed decision-making across various domains. Whereas, quantitative analysis using metrics such as BLEU and cosine similarity show that our approach outperforms the baseline by 10\%., Comment: 9 pages, ASONAM

This paper tackles the simultaneous optimization of pose and Neural Radiance Fields (NeRF). Departing from the conventional practice of using explicit global representations for camera pose, we propose a novel overparameterized representation that models camera poses as learnable rigid warp functions. We establish that modeling the rigid warps must be tightly coupled with constraints and regularization imposed. Specifically, we highlight the critical importance of enforcing invertibility when learning rigid warp functions via neural network and propose the use of an Invertible Neural Network (INN) coupled with a geometry-informed constraint for this purpose. We present results on synthetic and real-world datasets, and demonstrate that our approach outperforms existing baselines in terms of pose estimation and high-fidelity reconstruction due to enhanced optimization convergence., Comment: Accepted to ECCV 2024. Project page: https://sfchng.github.io/ineurowarping-github.io/

AI Agents are changing the way work gets done, both in consumer and enterprise domains. However, the design patterns and architectures to build highly capable agents or multi-agent systems are still developing, and the understanding of the implication of various design choices and algorithms is still evolving. In this paper, we present our work on building a novel web agent, Agent-E \footnote{Our code is available at \url{https://github.com/EmergenceAI/Agent-E}}. Agent-E introduces numerous architectural improvements over prior state-of-the-art web agents such as hierarchical architecture, flexible DOM distillation and denoising method, and the concept of \textit{change observation} to guide the agent towards more accurate performance. We first present the results of an evaluation of Agent-E on WebVoyager benchmark dataset and show that Agent-E beats other SOTA text and multi-modal web agents on this benchmark in most categories by 10-30\%. We then synthesize our learnings from the development of Agent-E into general design principles for developing agentic systems. These include the use of domain-specific primitive skills, the importance of distillation and de-noising of environmental observations, the advantages of a hierarchical architecture, and the role of agentic self-improvement to enhance agent efficiency and efficacy as the agent gathers experience.

Long-period radio transients are an emerging class of extreme astrophysical events of which only three are known. These objects emit highly polarised, coherent pulses of typically a few tens of seconds duration and minutes to hour-long periods. While magnetic white dwarfs and magnetars, either isolated or in binary systems, have been invoked to explain these objects, a consensus has not emerged. Here we report on the discovery of ASKAP J193505.1+214841.0 (henceforth ASKAPJ1935+2148) with a period of 53.8 minutes exhibiting three distinct emission states - a bright pulse state with highly linearly polarised pulses with widths of 10-50 seconds; a weak pulse state which is about 26 times fainter than the bright state with highly circularly polarised pulses of widths of approximately 370 milliseconds; and a quiescent or quenched state with no pulses. The first two states have been observed to progressively evolve over the course of 8 months with the quenched state interspersed between them suggesting physical changes in the region producing the emission. A constraint on the radius of the source for the observed period rules out a magnetic white dwarf origin. Unlike other long-period sources, ASKAPJ1935+2148 is the first to exhibit drastic variations in emission modes reminiscent of neutron stars. However, its radio properties challenge our current understanding of neutron star emission and evolution., Comment: Published in Nature Astronomy

Artificial intelligence and machine learning frameworks have served as computationally efficient mapping between inputs and outputs for engineering problems. These mappings have enabled optimization and analysis routines that have warranted superior designs, ingenious material systems and optimized manufacturing processes. A common occurrence in such modeling endeavors is the existence of multiple source of data, each differentiated by fidelity, operating conditions, experimental conditions, and more. Data fusion frameworks have opened the possibility of combining such differentiated sources into single unified models, enabling improved accuracy and knowledge transfer. However, these frameworks encounter limitations when the different sources are heterogeneous in nature, i.e., not sharing the same input parameter space. These heterogeneous input scenarios can occur when the domains differentiated by complexity, scale, and fidelity require different parametrizations. Towards addressing this void, a heterogeneous multi-source data fusion framework is proposed based on input mapping calibration (IMC) and latent variable Gaussian process (LVGP). In the first stage, the IMC algorithm is utilized to transform the heterogeneous input parameter spaces into a unified reference parameter space. In the second stage, a multi-source data fusion model enabled by LVGP is leveraged to build a single source-aware surrogate model on the transformed reference space. The proposed framework is demonstrated and analyzed on three engineering case studies (design of cantilever beam, design of ellipsoidal void and modeling properties of Ti6Al4V alloy). The results indicate that the proposed framework provides improved predictive accuracy over a single source model and transformed but source unaware model., Comment: 20 Pages,9 Figures, Data is available per request

Large Language Models such as GPTs (Generative Pre-trained Transformers) exhibit remarkable capabilities across a broad spectrum of applications. Nevertheless, due to their intrinsic complexity, these models present substantial challenges in interpreting their internal decision-making processes. This lack of transparency poses critical challenges when it comes to their adaptation by financial institutions, where concerns and accountability regarding bias, fairness, and reliability are of paramount importance. Mechanistic interpretability aims at reverse engineering complex AI models such as transformers. In this paper, we are pioneering the use of mechanistic interpretability to shed some light on the inner workings of large language models for use in financial services applications. We offer several examples of how algorithmic tasks can be designed for compliance monitoring purposes. In particular, we investigate GPT-2 Small's attention pattern when prompted to identify potential violation of Fair Lending laws. Using direct logit attribution, we study the contributions of each layer and its corresponding attention heads to the logit difference in the residual stream. Finally, we design clean and corrupted prompts and use activation patching as a causal intervention method to localize our task completion components further. We observe that the (positive) heads $10.2$ (head $2$, layer $10$), $10.7$, and $11.3$, as well as the (negative) heads $9.6$ and $10.6$ play a significant role in the task completion.

Haystack and Owens Valley Radio Observatory (OVRO) observations recently revealed strong sinusoidal total flux density variations that maintained coherence between 1975 and 2021 in the blazar PKS 2131-021 ($z=1.283)$. This was interpreted as possible evidence of a supermassive black hole binary (SMBHB). Extended observations through 2023 show coherence over 47.9~years, with an observed period $P_\textrm{15 GHz}=(1739.3 \pm 1.2) \, {\rm days}$. We reject, with $p$-value = $5.3 \times 10^{-7}$, the hypothesis that the variations are due to random fluctuations in the red noise tail of the power spectral density. There is clearly a constant-period physical phenomenon in PKS 2131-021 producing coherent intermittent sinusoidal flux density variations. We find the coherent sinusoidal intensity variations extend from below 2.7 GHz to optical frequencies, from which we derive an observed period $P_\textrm{optical}=(1764 \pm 36)$ days. Across this broad frequency range there is a monotonic phase shift in the sinusoidal variations with frequency. The same coherent periodicity is possibly also observed at $\gamma$-ray energies. The importance of well-vetted SMBHB candidates to searches for gravitational waves is pointed out. We estimate the fraction of blazars that are SMBHB candidates to be $>1$ in 100. Thus monitoring programs covering tens of thousands of blazars could discover hundreds of SMBHB candidates., Comment: 23 pages, 17 figures, 5 tables, submitted to ApJ

We study the spectral properties of SU(3) gauge theory with and without dynamical quarks (QCD) at thermal equilibrium using lattice gauge theory techniques. By measuring eigenstates of a massless overlap Dirac operator on the gauge configurations, we provide a gauge invariant method to study spectral properties of non-Abelian gauge theories. Whereas the majority of these eigenstates below the magnetic scale have universal nearest-neighbor level spacing fluctuations consistent with certain class of random matrix theories at temperatures away from the chiral crossover transition in QCD, a few among them start to become prominent just above the crossover forming clusters percolating over the entire volume. By matching the non-perturbative magnetic scales in a high temperature thermal state and a particular non-equilibrium chaotic state of QCD, we provide an estimate of thermalization time $\sim 1.44$ fm/c., Comment: 7 pages, 4 figures

The development of a colloidal synthesis procedure to produce nanomaterials of a specific size with high shape and size purity is often a time consuming, iterative process. This is often due to the time, resource and expertise intensive characterization methods required for quantitative determination of nanomaterial size and shape. Absorption spectroscopy is often the easiest method of colloidal nanomaterial characterization, however, due to the lack of a reliable method to extract nanoparticle shapes from absorption spectroscopy, it is generally treated as a more qualitative measure for metal nanoparticles. This work demonstrates a gold nanorod (AuNR) spectral morphology analysis (SMA) tool, AuNR-SMA, which is a fast and accurate method to extract quantitative information about an AuNR sample's structural parameters from its absorption spectra. We apply AuNR-SMA in three distinct applications. First, we demonstrate its utility as an automated analysis tool in a high throughput AuNR synthesis procedure by generating quantitative size information from optical spectra. Second, we use the predictions generated by this model to train a machine learning model capable of predicting the resulting AuNR size distributions from the reaction conditions used to synthesize them. Third, we turn this model to spectra extracted from the literature where no size distributions are reported to impute unreported quantitative information of AuNR synthesis. This approach can potentially be extended to any other nanocrystal system where the absorption spectra are size dependent and accurate numerical simulation of the absorption spectra is possible. In addition, this pipeline could be integrated into automated synthesis apparatuses to provide interpretable data from simple measurements and help explore the synthesis science of nanoparticles in a rational manner or facilitate closed-loop workflows.

Retrieval Augmented Generation (RAG) techniques aim to mitigate hallucinations in Large Language Models (LLMs). However, LLMs can still produce information that is unsupported or contradictory to the retrieved contexts. We introduce LYNX, a SOTA hallucination detection LLM that is capable of advanced reasoning on challenging real-world hallucination scenarios. To evaluate LYNX, we present HaluBench, a comprehensive hallucination evaluation benchmark, consisting of 15k samples sourced from various real-world domains. Our experiment results show that LYNX outperforms GPT-4o, Claude-3-Sonnet, and closed and open-source LLM-as-a-judge models on HaluBench. We release LYNX, HaluBench and our evaluation code for public access.

One of the most critical tasks of Microsoft sellers is to meticulously track and nurture potential business opportunities through proactive engagement and tailored solutions. Recommender systems play a central role to help sellers achieve their goals. In this paper, we present a content recommendation model which surfaces various types of content (technical documentation, comparison with competitor products, customer success stories etc.) that sellers can share with their customers or use for their own self-learning. The model operates at the opportunity level which is the lowest possible granularity and the most relevant one for sellers. It is based on semantic matching between metadata from the contents and carefully selected attributes of the opportunities. Considering the volume of seller-managed opportunities in organizations such as Microsoft, we show how to perform efficient semantic matching over a very large number of opportunity-content combinations. The main challenge is to ensure that the top-5 relevant contents for each opportunity are recommended out of a total of $\approx 40,000$ published contents. We achieve this target through an extensive comparison of different model architectures and feature selection. Finally, we further examine the quality of the recommendations in a quantitative manner using a combination of human domain experts as well as by using the recently proposed "LLM as a judge" framework.

The accurate evaluation of left atrial fibrosis via high-quality 3D Late Gadolinium Enhancement (LGE) MRI is crucial for atrial fibrillation management but is hindered by factors like patient movement and imaging variability. The pursuit of automated LGE MRI quality assessment is critical for enhancing diagnostic accuracy, standardizing evaluations, and improving patient outcomes. The deep learning models aimed at automating this process face significant challenges due to the scarcity of expert annotations, high computational costs, and the need to capture subtle diagnostic details in highly variable images. This study introduces HAMIL-QA, a multiple instance learning (MIL) framework, designed to overcome these obstacles. HAMIL-QA employs a hierarchical bag and sub-bag structure that allows for targeted analysis within sub-bags and aggregates insights at the volume level. This hierarchical MIL approach reduces reliance on extensive annotations, lessens computational load, and ensures clinically relevant quality predictions by focusing on diagnostically critical image features. Our experiments show that HAMIL-QA surpasses existing MIL methods and traditional supervised approaches in accuracy, AUROC, and F1-Score on an LGE MRI scan dataset, demonstrating its potential as a scalable solution for LGE MRI quality assessment automation. The code is available at: $\href{https://github.com/arf111/HAMIL-QA}{\text{this https URL}}$, Comment: Accepted to MICCAI2024, 10 pages, 2 figures

This paper introduces the concept of the closed neighborhood corona product of the graph. We explore the mathematical features of this product graph, specifically in terms of its spectral characteristics. We have calculated the characteristic polynomials of the adjacency, Laplacian, and signless Laplacian matrices. Moreover, we investigate the conditions under which two graphs are cospectral regarding this product. A significant portion of our study is dedicated to computing the Kirchhoff index, the number of spanning trees and the sequence of non-cospectral equienergetic product graphs. We also outline specific criteria that determine when the product graph is integral.

It is desirable to coarsely classify short scientific texts, such as grant or publication abstracts, for strategic insight or research portfolio management. These texts efficiently transmit dense information to experts possessing a rich body of knowledge to aid interpretation. Yet this task is remarkably difficult to automate because of brevity and the absence of context. To address this gap, we have developed a novel approach to generate and appropriately assign coarse domain-specific labels. We show that a Large Language Model (LLM) can provide metadata essential to the task, in a process akin to the augmentation of supplemental knowledge representing human intuition, and propose a workflow. As a pilot study, we use a corpus of award abstracts from the National Aeronautics and Space Administration (NASA). We develop new assessment tools in concert with established performance metrics.

The $q-$state clock model, sometimes called the discrete $XY$ model, is known to show a second-order (symmetry breaking) phase transition in two-dimension (2D) for $q\le 4$ ($q=2$ corresponds to the Ising model). On the other hand, the $q\to\infty$ limit of the model corresponds to the $XY$ model, which shows the infinite order (non-symmetry breaking) Berezinskii-Kosterlitz-Thouless (BKT) phase transition in 2D. Interestingly, the 2D clock model with $q\ge 5$ is predicted to show three different phases and two associated phase transitions. There are varying opinions about the actual characters of phases and the associated transitions. In this work, we develop the basic and higher-order mean-field (MF) theories to study the $q$-state clock model systematically. Our MF calculations reaffirm that, for large $q$, there are three phases: (broken) $\mathbb{Z}_q$ symmetric ferromagnetic phase at the low temperature, emergent $U(1)$ symmetric BKT phase at the intermediate temperature, and paramagnetic (disordered) phase at the high temperature. The phase transition at the higher temperature is found to be of the BKT type, and the other transition at the lower temperature is argued to be a large-order spontaneous symmetry-breaking (SSB) type (the largeness of transition order yields the possibility of having some of the numerical characteristics of a BKT transition). The higher-order MF theory developed here better characterizes phases by estimating the spin-spin correlation between two neighbors., Comment: Some new results added, parts of text rewritten, 16 pages, 9 figures

We consider the problem of communicating over a channel that breaks the message block into fragments of random lengths, shuffles them out of order, and deletes a random fraction of the fragments. Such a channel is motivated by applications in molecular data storage and forensics, and we refer to it as the torn-paper channel. We characterize the capacity of this channel under arbitrary fragment length distributions and deletion probabilities. Precisely, we show that the capacity is given by a closed-form expression that can be interpreted as F - A, where F is the coverage fraction ,i.e., the fraction of the input codeword that is covered by output fragments, and A is an alignment cost incurred due to the lack of ordering in the output fragments. We then consider a noisy version of the problem, where the fragments are corrupted by binary symmetric noise. We derive upper and lower bounds to the capacity, both of which can be seen as F - A expressions. These bounds match for specific choices of fragment length distributions, and they are approximately tight in cases where there are not too many short fragments., Comment: 43 pages, 3 figures

The Baryon Acoustic Oscillation (BAO) feature in the two-point correlation function (TPCF) of discrete tracers such as galaxies is an accurate standard ruler. The covariance matrix of the TPCF plays an important role in determining how the precision of this ruler depends on the number density and clustering strength of the tracers, as well as the survey volume. An eigen-decomposition of this matrix provides an objective way to separate the contributions of cosmic variance from those of shot-noise to the statistical uncertainties. For the signal-to-noise levels that are expected in ongoing and next-generation surveys, the cosmic variance eigen-modes dominate. These modes are smooth functions of scale, meaning that: they are insensitive to the modest changes in binning that are allowed if one wishes to resolve the BAO feature in the TPCF; they provide a good description of the correlated residuals which result from fitting smooth functional forms to the measured TPCF; they motivate a simple but accurate approximation for the uncertainty on the Linear Point (LP) estimate of the BAO distance scale. This approximation allows one to quantify the precision of the BAO distance scale estimate without having to generate a large ensemble of mock catalogs and explains why: the uncertainty on the LP does not depend on the functional form fitted to the TPCF or the binning used; the LP is more constraining than the peak or dip scales in the TPCF; the evolved TPCF is less constraining than the initial one, so that reconstruction schemes can yield significant gains in precision., Comment: 11 pages, 9 figures, accepted by Physical Review D

Certifiable robustness gives the guarantee that small perturbations around an input to a classifier will not change the prediction. There are two approaches to provide certifiable robustness to adversarial examples: a) explicitly training classifiers with small Lipschitz constants, and b) Randomized smoothing, which adds random noise to the input to create a smooth classifier. We propose \textit{SPLITZ}, a practical and novel approach which leverages the synergistic benefits of both the above ideas into a single framework. Our main idea is to \textit{split} a classifier into two halves, constrain the Lipschitz constant of the first half, and smooth the second half via randomization. Motivation for \textit{SPLITZ} comes from the observation that many standard deep networks exhibit heterogeneity in Lipschitz constants across layers. \textit{SPLITZ} can exploit this heterogeneity while inheriting the scalability of randomized smoothing. We present a principled approach to train \textit{SPLITZ} and provide theoretical analysis to derive certified robustness guarantees during inference. We present a comprehensive comparison of robustness-accuracy tradeoffs and show that \textit{SPLITZ} consistently improves upon existing state-of-the-art approaches on MNIST and CIFAR-10 datasets. For instance, with $\ell_2$ norm perturbation budget of \textbf{$\epsilon=1$}, \textit{SPLITZ} achieves $\textbf{43.2\%}$ top-1 test accuracy on CIFAR-10 dataset compared to state-of-art top-1 test accuracy $\textbf{39.8\%}

This work reports an acoustic solidly mounted resonator (SMR) at 18.64 GHz, among the highest operating frequencies reported. The device is built in scandium aluminum nitride (ScAlN) on top of silicon dioxide (SiO2) and tantalum pentoxide (Ta2O5) Bragg reflectors on silicon (Si) wafer. The stack is analyzed with X-ray reflectivity (XRR) and high-resolution X-ray diffraction (HRXRD). The resonator shows a coupling coefficient (k2) of 2.0%, high series quality factor (Qs) of 156, shunt quality factor (Qp) of 142, and maximum Bode quality factor (Qmax) of 210. The third-order harmonics at 59.64 GHz is also observed with k2 around 0.6% and Q around 40. Upon further development, the reported acoustic resonator platform can enable various front-end signal-processing functions, e.g., filters and oscillators, at future frequency range 3 (FR3) bands., Comment: 5 pages, 9 figures, 5 tables

Advancements in Large Language Models (LLMs) have significantly enhanced instruction-following capabilities. However, most Instruction Fine-Tuning (IFT) datasets are predominantly in English, limiting model performance in other languages. Traditional methods for creating multilingual IFT datasets such as translating existing English IFT datasets or converting existing NLP datasets into IFT datasets by templating, struggle to capture linguistic nuances and ensure prompt (instruction) diversity. To address this issue, we propose a novel method for collecting multilingual IFT datasets that preserves linguistic naturalness and ensures prompt diversity. This approach leverages English-focused LLMs, monolingual corpora, and a scoring function to create high-quality, diversified IFT datasets in multiple languages. Experiments demonstrate that LLMs finetuned using these IFT datasets show notable improvements in both generative and discriminative tasks, indicating enhanced language comprehension by LLMs in non-English contexts. Specifically, on the multilingual summarization task, LLMs using our IFT dataset achieved 17.57% and 15.23% improvements over LLMs fine-tuned with translation-based and template-based datasets, respectively.

Despite recent advancements in vision-language models, their performance remains suboptimal on images from non-western cultures due to underrepresentation in training datasets. Various benchmarks have been proposed to test models' cultural inclusivity, but they have limited coverage of cultures and do not adequately assess cultural diversity across universal as well as culture-specific local concepts. To address these limitations, we introduce the GlobalRG benchmark, comprising two challenging tasks: retrieval across universals and cultural visual grounding. The former task entails retrieving culturally diverse images for universal concepts from 50 countries, while the latter aims at grounding culture-specific concepts within images from 15 countries. Our evaluation across a wide range of models reveals that the performance varies significantly across cultures -- underscoring the necessity for enhancing multicultural understanding in vision-language models., Comment: Under peer review

"Garbage In Garbage Out" is a universally agreed quote by computer scientists from various domains, including Artificial Intelligence (AI). As data is the fuel for AI, models trained on low-quality, biased data are often ineffective. Computer scientists who use AI invest a considerable amount of time and effort in preparing the data for AI. However, there are no standard methods or frameworks for assessing the "readiness" of data for AI. To provide a quantifiable assessment of the readiness of data for AI processes, we define parameters of AI data readiness and introduce AIDRIN (AI Data Readiness Inspector). AIDRIN is a framework covering a broad range of readiness dimensions available in the literature that aid in evaluating the readiness of data quantitatively and qualitatively. AIDRIN uses metrics in traditional data quality assessment such as completeness, outliers, and duplicates for data evaluation. Furthermore, AIDRIN uses metrics specific to assess data for AI, such as feature importance, feature correlations, class imbalance, fairness, privacy, and FAIR (Findability, Accessibility, Interoperability, and Reusability) principle compliance. AIDRIN provides visualizations and reports to assist data scientists in further investigating the readiness of data. The AIDRIN framework enhances the efficiency of the machine learning pipeline to make informed decisions on data readiness for AI applications., Comment: 12 pages, 9 figures, Accepted to SSDBM 2024

The scaling of large language models to encode all the world's knowledge in model parameters is unsustainable and has exacerbated resource barriers. Retrieval-Augmented Generation (RAG) presents a potential solution, yet its application to vision-language models (VLMs) is under explored. Existing methods focus on models designed for single tasks. Furthermore, they're limited by the need for resource intensive pre training, additional parameter requirements, unaddressed modality prioritization and lack of clear benefit over non-retrieval baselines. This paper introduces RAVEN, a multitask retrieval augmented VLM framework that enhances base VLMs through efficient, task specific fine-tuning. By integrating retrieval augmented samples without the need for additional retrieval-specific parameters, we show that the model acquires retrieval properties that are effective across multiple tasks. Our results and extensive ablations across retrieved modalities for the image captioning and VQA tasks indicate significant performance improvements compared to non retrieved baselines +1 CIDEr on MSCOCO, +4 CIDEr on NoCaps and nearly a +3\% accuracy on specific VQA question types. This underscores the efficacy of applying RAG approaches to VLMs, marking a stride toward more efficient and accessible multimodal learning.

We study the differentially private (DP) empirical risk minimization (ERM) problem under the semi-sensitive DP setting where only some features are sensitive. This generalizes the Label DP setting where only the label is sensitive. We give improved upper and lower bounds on the excess risk for DP-ERM. In particular, we show that the error only scales polylogarithmically in terms of the sensitive domain size, improving upon previous results that scale polynomially in the sensitive domain size (Ghazi et al., 2021)., Comment: To appear in COLT 2024

We engineer blockchain based risk managed portfolios by creating three funds with distinct risk and return profiles: 1) Alpha - high risk portfolio; 2) Beta - mimics the wider market; and 3) Gamma - represents the risk free rate adjusted to beat inflation. Each of the sub-funds (Alpha, Beta and Gamma) provides risk parity because the weight of each asset in the corresponding portfolio is set to be inversely proportional to the risk derived from investing in that asset. This can be equivalently stated as equal risk contributions from each asset towards the overall portfolio risk. We provide detailed mechanics of combining assets - including mathematical formulations - to obtain better risk managed portfolios. The descriptions are intended to show how a risk parity based efficient frontier portfolio management engine - that caters to different risk appetites of investors by letting each individual investor select their preferred risk-return combination - can be created seamlessly on blockchain. Any Investor - using decentralized ledger technology - can select their desired level of risk, or return, and allocate their wealth accordingly among the sub funds, which balance one another under different market conditions. This evolution of the risk parity principle - resulting in a mechanism that is geared to do well under all market cycles - brings more robust performance and can be termed as conceptual parity. We have given several numerical examples that illustrate the various scenarios that arise when combining Alpha, Beta and Gamma to obtain Parity. The final investment frontier is now possible - a modification to the efficient frontier, thus becoming more than a mere theoretical construct - on blockchain since anyone from anywhere can participate at anytime to obtain wealth appreciation based on their financial goals.

We study the problem of computing pairwise statistics, i.e., ones of the form $\binom{n}{2}^{-1} \sum_{i \ne j} f(x_i, x_j)$, where $x_i$ denotes the input to the $i$th user, with differential privacy (DP) in the local model. This formulation captures important metrics such as Kendall's $\tau$ coefficient, Area Under Curve, Gini's mean difference, Gini's entropy, etc. We give several novel and generic algorithms for the problem, leveraging techniques from DP algorithms for linear queries., Comment: Published in NeurIPS 2023

Pixel-level Video Understanding in the Wild Challenge (PVUW) focus on complex video understanding. In this CVPR 2024 workshop, we add two new tracks, Complex Video Object Segmentation Track based on MOSE dataset and Motion Expression guided Video Segmentation track based on MeViS dataset. In the two new tracks, we provide additional videos and annotations that feature challenging elements, such as the disappearance and reappearance of objects, inconspicuous small objects, heavy occlusions, and crowded environments in MOSE. Moreover, we provide a new motion expression guided video segmentation dataset MeViS to study the natural language-guided video understanding in complex environments. These new videos, sentences, and annotations enable us to foster the development of a more comprehensive and robust pixel-level understanding of video scenes in complex environments and realistic scenarios. The MOSE challenge had 140 registered teams in total, 65 teams participated the validation phase and 12 teams made valid submissions in the final challenge phase. The MeViS challenge had 225 registered teams in total, 50 teams participated the validation phase and 5 teams made valid submissions in the final challenge phase., Comment: MOSE Challenge: https://henghuiding.github.io/MOSE/ChallengeCVPR2024, MeViS Challenge: https://henghuiding.github.io/MeViS/ChallengeCVPR2024

Conventional rendering techniques are primarily designed and optimized for single-frame rendering. In practical applications, such as scene editing and animation rendering, users frequently encounter scenes where only a small portion is modified between consecutive frames. In this paper, we develop a novel approach to incremental re-rendering of scenes with dynamic objects, where only a small part of a scene moves from one frame to the next. We formulate the difference (or residual) in the image between two frames as a (correlated) light-transport integral which we call the residual path integral. Efficient numerical solution of this integral then involves (1)~devising importance sampling strategies to focus on paths with non-zero residual-transport contributions and (2)~choosing appropriate mappings between the native path spaces of the two frames. We introduce a set of path importance sampling strategies that trace from the moving object(s) which are the sources of residual energy. We explore path mapping strategies that generalize those from gradient-domain path tracing to our importance sampling techniques specially for dynamic scenes. Additionally, our formulation can be applied to material editing as a simpler special case. We demonstrate speed-ups over previous correlated sampling of path differences and over rendering the new frame independently. Our formulation brings new insights into the re-rendering problem and paves the way for devising new types of sampling techniques and path mappings with different trade-offs., Comment: 14 pages, 13 figures

Large language models (LLMs) are typically multilingual due to pretraining on diverse multilingual corpora. But can these models relate corresponding concepts across languages, effectively being crosslingual? This study evaluates six state-of-the-art LLMs on inherently crosslingual tasks. We observe that while these models show promising surface-level crosslingual abilities on machine translation and embedding space analyses, they struggle with deeper crosslingual knowledge transfer, revealing a crosslingual knowledge barrier in both general (MMLU benchmark) and domain-specific (Harry Potter quiz) contexts. We observe that simple inference-time mitigation methods offer only limited improvement. On the other hand, we propose fine-tuning of LLMs on mixed-language data, which effectively reduces these gaps, even when using out-of-domain datasets like WikiText. Our findings suggest the need for explicit optimization to unlock the full crosslingual potential of LLMs. Our code is publicly available at https://github.com/google-research/crosslingual-knowledge-barriers.

Variational quantum algorithms (VQAs) show potential for quantum advantage in the near term of quantum computing, but demand a level of accuracy that surpasses the current capabilities of NISQ devices. To systematically mitigate the impact of quantum device error on VQAs, we propose Clapton: Clifford-Assisted Problem Transformation for Error Mitigation in Variational Quantum Algorithms. Clapton leverages classically estimated good quantum states for a given VQA problem, classical simulable models of device noise, and the variational principle for VQAs. It applies transformations on the VQA problem's Hamiltonian to lower the energy estimates of known good VQA states in the presence of the modeled device noise. The Clapton hypothesis is that as long as the known good states of the VQA problem are close to the problem's ideal ground state and the device noise modeling is reasonably accurate (both of which are generally true), then the Clapton transformation substantially decreases the impact of device noise on the ground state of the VQA problem, thereby increasing the accuracy of the VQA solution. Clapton is built as an end-to-end application-to-device framework and achieves mean VQA initialization improvements of 1.7x to 3.7x, and up to a maximum of 13.3x, over the state-of-the-art baseline when evaluated for a variety of scientific applications from physics and chemistry on noise models and real quantum devices.

Identifying patient cohorts is fundamental to numerous healthcare tasks, including clinical trial recruitment and retrospective studies. Current cohort retrieval methods in healthcare organizations rely on automated queries of structured data combined with manual curation, which are time-consuming, labor-intensive, and often yield low-quality results. Recent advancements in large language models (LLMs) and information retrieval (IR) offer promising avenues to revolutionize these systems. Major challenges include managing extensive eligibility criteria and handling the longitudinal nature of unstructured Electronic Medical Records (EMRs) while ensuring that the solution remains cost-effective for real-world application. This paper introduces a new task, Automatic Cohort Retrieval (ACR), and evaluates the performance of LLMs and commercial, domain-specific neuro-symbolic approaches. We provide a benchmark task, a query dataset, an EMR dataset, and an evaluation framework. Our findings underscore the necessity for efficient, high-quality ACR systems capable of longitudinal reasoning across extensive patient databases.

Curiously in the minimal left right symmetric model, chiral symmetry that protects the electron's mass (m_e), due to parity (P) implies the vanishing of its neutrino mixing angles. We break this symmetry softly (or spontaneously if it is gauged) to generate the observed large neutrino mixing angles at the tree-level. The electron then acquires its mass on RGE running due to its neutrino's mixing, and in turn determines the B-L gauge symmetry breaking scale (v_R) to be $10^{10} GeV \lesssim v_R \leq 10^{15} GeV. $ If the muon's mass is also generated radiatively, the B-L breaking scale is $\sim 10^{14-15}$ GeV. Regardless of the high scale of v_R, this is a testable model since on RGE running and P breaking, a large strong CP phase ($\bar{\theta} >> 10^{-10}$) which depends logarithmically on $v_R$ is generated if there is $\mathcal{O}(1)$ CP violation in leptonic Yukawa couplings. Hence we expect that leptonic CP phases including the Dirac CP phase $\delta_{CP}$ of the PMNS matrix must be consistent with 0 or 180^o to within a degree, which can be verified or excluded by neutrino experiments such as DUNE and Hyper-Kamiokande. In lieu of P, if charge conjugation C is used, the same results follow. However with C and no P, axions would likely need to be added anyway, in which case there is no constraint on \delta_{CP}., Comment: Added a section that gauges the U(1)_L x U(1)_R global symmetry that sets the tree level electron Yukawa to zero

Large language models (LLMs) have emerged as powerful tools for tackling complex tasks across diverse domains, but they also raise privacy concerns when fine-tuned on sensitive data due to potential memorization. While differential privacy (DP) offers a promising solution by ensuring models are 'almost indistinguishable' with or without any particular privacy unit, current evaluations on LLMs mostly treat each example (text record) as the privacy unit. This leads to uneven user privacy guarantees when contributions per user vary. We therefore study user-level DP motivated by applications where it necessary to ensure uniform privacy protection across users. We present a systematic evaluation of user-level DP for LLM fine-tuning on natural language generation tasks. Focusing on two mechanisms for achieving user-level DP guarantees, Group Privacy and User-wise DP-SGD, we investigate design choices like data selection strategies and parameter tuning for the best privacy-utility tradeoff., Comment: Published as a conference paper at COLM 2024