Search

We lack a systematic understanding of the effects of fine-tuning (via methods such as instruction-tuning or reinforcement learning from human feedback), particularly on tasks outside the narrow fine-tuning distribution. In a simplified scenario, we demonstrate that improving performance on tasks within the fine-tuning data distribution comes at the expense of capabilities on other tasks. We hypothesize that language models implicitly infer the task of the prompt and that fine-tuning skews this inference towards tasks in the fine-tuning distribution. To test this, we propose Conjugate Prompting, which artificially makes the task look farther from the fine-tuning distribution while requiring the same capability, and we find that this recovers some of the pretraining capabilities in our synthetic setup. Since real-world fine-tuning distributions are predominantly English, we apply conjugate prompting to recover pretrained capabilities in LLMs by simply translating the prompts to different languages. This allows us to recover in-context learning abilities lost via instruction tuning, natural reasoning capability lost during code fine-tuning, and, more concerningly, harmful content generation suppressed by safety fine-tuning in chatbots like ChatGPT., Comment: ICLR 2024

Recently, the region surrounding eHWC J1842-035 has been studied extensively by gamma-ray observatories due to its extended emission reaching up to a few hundred TeV and potential as a hadronic accelerator. In this work, we use 1,910 days of cumulative data from the High Altitude Water Cherenkov (HAWC) observatory to carry out a dedicated systematic source search of the eHWC J1842-035 region. During the search we have found three sources in the region, namely, HAWC J1844-034, HAWC J1843-032, and HAWC J1846-025. We have identified HAWC J1844-034 as the extended source that emits photons with energies up to 175 TeV. We compute the spectrum for HAWC J1844-034 and by comparing with the observational results from other experiments, we have identified HESS J1843-033, LHAASO J1843-0338, and TASG J1844-038 as very-high-energy gamma-ray sources with a matching origin. Also, we present and use the multi-wavelength data to fit the hadronic and leptonic particle spectra. We have identified four pulsar candidates in the nearby region from which PSR J1844-0346 is found to be the most likely candidate due to its proximity to HAWC J1844-034 and the computed energy budget. We have also found SNR G28.6-0.1 as a potential counterpart source of HAWC J1844-034 for which both leptonic and hadronic scenarios are feasible., Comment: 13 pages, 9 figures, published in ApJ

The decay or annihilation of dark matter particles may produce a steady flux of very-high-energy gamma rays detectable above the diffuse background. Nearby clusters of galaxies provide excellent targets to search for the signatures of particle dark matter interactions. In particular, the Virgo cluster spans several degrees across the sky and can be efficiently probed with a wide field-of-view instrument. The High Altitude Water Cherenkov (HAWC) observatory, due to its wide field of view and sensitivity to gamma rays at an energy scale of 300 GeV--100 TeV is well-suited for this search. Using 2141 days of data, we search for gamma-ray emission from the Virgo cluster, assuming well-motivated dark matter sub-structure models. Our results provide some of the strongest constraints on the decay lifetime of dark matter for masses above 10 TeV., Comment: 7 pages, 3 figures. Accepted for publication in Physical Review D

We consider normal rational projective surfaces with torus action and provide a formula for their Picard index, that means the index of the Picard group inside the divisor class group. As an application, we classify the log del Pezzo surfaces with torus action of Picard number one up to Picard index 10,000., Comment: 30 pages, 3 figures

We present the NVIDIA cuQuantum SDK, a state-of-the-art library of composable primitives for GPU-accelerated quantum circuit simulations. As the size of quantum devices continues to increase, making their classical simulation progressively more difficult, the availability of fast and scalable quantum circuit simulators becomes vital for quantum algorithm developers, as well as quantum hardware engineers focused on the validation and optimization of quantum devices. The cuQuantum SDK was created to accelerate and scale up quantum circuit simulators developed by the quantum information science community by enabling them to utilize efficient scalable software building blocks optimized for NVIDIA GPU platforms. The functional building blocks provided cover the needs of both state vector- and tensor network- based simulators, including approximate tensor network simulation methods based on matrix product state, projected entangled pair state, and other factorized tensor representations. By leveraging the enormous computing power of the latest NVIDIA GPU architectures, quantum circuit simulators that have adopted the cuQuantum SDK demonstrate significant acceleration, compared to CPU-only execution, for both the state vector and tensor network simulation methods. Furthermore, by utilizing the parallel primitives available in the cuQuantum SDK, one can easily transition to distributed GPU-accelerated platforms, including those furnished by cloud service providers and high-performance computing systems deployed by supercomputing centers, extending the scale of possible quantum circuit simulations. The rich capabilities provided by the SDK are conveniently made available via both Python and C application programming interfaces, where the former is directly targeting a broad Python quantum community and the latter allows tight integration with simulators written in any programming language., Comment: paper accepted at QCE 2023, journal reference will be updated whenever available

Background: Research has consistently shown that disadvantaged students have inequitable access to high-quality teaching (e.g., Isenberg et al., 2013; Goldhaber et al. 2018; Hanselman, 2019). This non-random sorting of students to teachers has been documented at multiple organizational levels, including districts, schools and classrooms (e.g., Goldhaber et al., 2015; Clotfelter et al., 2023). Studies that have compared sorting across the different levels have shown that classroom-level inequalities tend to be the smallest in magnitude. However, classroom-level sorting remains a potentially high-impact target for policy intervention for several reasons. First, the mechanism through which classroom-level sorting occurs (i.e., classroom rostering) is under the direct purview of school staff and therefore highly malleable. Second, schools already have the resources required to implement classroom rostering, so changes to rostering policies have the potential to be cost-neutral or even cost-efficient. Third, because every student is rostered each year, rostering policies represent a "high volume, low margin" approach to producing net gains in equitable access and student learning. Purpose and Research Questions: The present study is a secondary analysis of administrative data obtained from the North Carolina Department of Public Instruction (NC DPI). Using this data source, we seek to more fully understand the potential of classroom rostering to reduce inequitable access to high-quality teaching and improve student learning. We approach this overall goal in two main steps. In the first step, we descriptively analyze the assignment of students to teachers within school- grade-year cells. Here we are interested to better understand the student characteristics that are associated with differential access to high-quality teaching. We categorize teaching quality using the tertiles of the distribution of teachers' value-added (VA) in the previous school year, and treat teachers without prior VA as a fourth category. We address the following descriptive questions: 1a) What proportion of school-grade-year cells have sufficient variability in effective teaching for rostering to be a potential mechanism for inequitable access to high-quality teaching? 1b) Which, if any, student characteristics are associated with differential assignment to teachers of different VA categories? Step 2 of the study builds on Step 1 by using the results of Research Question 1b to identify two contrasting subgroups of schools, one group with relatively equitable rostering practices and the other with relatively inequitable rostering practices. In these two contrasting cases, we address two additional questions about the effects of each of three counterfactual rostering policies. The three policies are: Counterfactual Policy 1: Balanced classrooms (no tracking). Counterfactual Policy 2: Assign stronger teachers to students with lower prior achievement. Counterfactual Policy 3: Assign students to teachers to maximize net predicted learning gains while maintaining balance. Clotfelter et al. (2023) recently suggested the first two policies and the third is novel to this study. We study the effects of each policy on two outcomes of interest: 2a) access to high-quality teaching and 2b) student learning. Data Collection and Analysis. We conduct a secondary analysis of administrative data from NC DPI over the 2011-2012 to 2018-2019 school years. We focus on ELA and Math instruction for students in 4th and 5th grades. The panel includes data on 1.2 million students and 57,854 teachers across 1730 schools from all 110 districts in NC. Data cleaning was completed in early Spring 2024, with a working paper expected to be completed in Summer 2024. Step 1. To address Research Question 1a we tabulate the number of teachers in each VA category within school-grade-year cells. To address Research Question 1b we use Poisson regression with fixed-effects for school-grade-year cells (see Springer et al., 2022). We consider a standard fixed effects specification and a specification that additionally incorporates random slopes on coefficients describing differential assignment. This is an extension of Mundlak's (1978) specification for panel data that allows us to study how rostering practices differ over school-grade-year cells, while still focusing on sources of variation that arise within cells. Step 2. We use the predicted random slopes from Research Question 1b to create two groups of schools, those in which there is strong evidence of highly equitable rostering, and those in which there is strong evidence of highly inequitable rostering. The goal is to identify 50-100 schools that unambiguously fall into each group and contrast results across groups. The three counterfactual rostering policies can be implemented by solving standard mixed- integer programs (MIPs) within each school-grade-year cell. Details of the objectives and constraints of these MIPs are currently under investigation and will be reported. The MIP solutions are a set of binary decision variables indicating whether each student is assigned to each teacher -- i.e., a set of counterfactual rosters. Having the counterfactual rosters in hand, Research Question 2a can be addressed by applying the same Poisson regression model described for Research Question 1b, but this time using the counterfactual rosters. It is important to note that Research Question 2b and Counterfactual Policy 3 both require specifying a predictive model for student learning that incorporates complementarity between students and teachers. In the present context, this predictive model is referred to as a match- output function (Graham, 2011). We have previously researched match-output functions for classroom rostering (Halpin et al. 2022) and arrived at a specification in which teachers' leave- year-out VA is estimated separately for each tertile of the within-school distribution of student prior achievement. This is a special case of Delgado's (2022) approach to teacher comparative advantage. We discuss the strengths and limitations of this specification, the latter mainly having to do with imprecise estimation of VA within prior achievement tertiles. We also discuss the conditions under which the match-output function is causally identified (Graham et al., 2016). Results are summarized in terms of the average re-allocation effects on student learning for each of the three counterfactual policies, computed within each school-grade-year cell. Research Question 2b will be addressed by comparing the distribution re-allocation effects in the two contrasting subgroups of schools.

Understanding the mechanisms of shock-induced pore collapse is of great interest in various disciplines in sciences and engineering, including materials science, biological sciences, and geophysics. However, numerical modeling of the complex pore collapse processes can be costly. To this end, a strong need exists to develop surrogate models for generating economic predictions of pore collapse processes. In this work, we study the use of a data-driven reduced order model, namely dynamic mode decomposition, and a deep generative model, namely conditional generative adversarial networks, to resemble the numerical simulations of the pore collapse process at representative training shock pressures. Since the simulations are expensive, the training data are scarce, which makes training an accurate surrogate model challenging. To overcome the difficulties posed by the complex physics phenomena, we make several crucial treatments to the plain original form of the methods to increase the capability of approximating and predicting the dynamics. In particular, physics information is used as indicators or conditional inputs to guide the prediction. In realizing these methods, the training of each dynamic mode composition model takes only around 30 seconds on CPU. In contrast, training a generative adversarial network model takes 8 hours on GPU. Moreover, using dynamic mode decomposition, the final-time relative error is around 0.3% in the reproductive cases. We also demonstrate the predictive power of the methods at unseen testing shock pressures, where the error ranges from 1.3% to 5% in the interpolatory cases and 8% to 9% in extrapolatory cases.

We here address the problem of fairly allocating indivisible goods or chores to $n$ agents with weights that define their entitlement to the set of indivisible resources. Stemming from well-studied fairness concepts such as envy-freeness up to one good (EF1) and envy-freeness up to any good (EFX) for agents with equal entitlements, we present, in this study, the first set of impossibility results alongside algorithmic guarantees for fairness among agents with unequal entitlements. Within this paper, we expand the concept of envy-freeness up to any good or chore to the weighted context (WEFX and XWEF respectively), demonstrating that these allocations are not guaranteed to exist for two or three agents. Despite these negative results, we develop a WEFX procedure for two agents with integer weights, and furthermore, we devise an approximate WEFX procedure for two agents with normalized weights. We further present a polynomial-time algorithm that guarantees a weighted envy-free allocation up to one chore (1WEF) for any number of agents with additive cost functions. Our work underscores the heightened complexity of the weighted fair division problem when compared to its unweighted counterpart., Comment: Appears in the 38th AAAI Conference on Artificial Intelligence (AAAI), 2024

The Galactic Halo is a key target for indirect dark matter detection. The High Altitude Water Cherenkov (HAWC) observatory is a high-energy (~300 GeV to >100 TeV) gamma-ray detector located in central Mexico. HAWC operates via the water Cherenkov technique and has both a wide field of view of 2 sr and a >95% duty cycle, making it ideal for analyses of highly extended sources. We made use of these properties of HAWC and a new background-estimation technique optimized for extended sources to probe a large region of the Galactic Halo for dark matter signals. With this approach, we set improved constraints on dark matter annihilation and decay between masses of 10 and 100 TeV. Due to the large spatial extent of the HAWC field of view, these constraints are robust against uncertainties in the Galactic dark matter spatial profile., Comment: 19 pages, 14 figures

We present a short overview of the TeV-Halos objects as a discovery and a relevant contribution of the High Altitude Water \v{C}erenkov (HAWC) observatory to TeV astrophysics. We discuss history, discovery, knowledge, and the next step through a new and more detailed analysis than the original study in 2017. TeV-Halos will contribute to resolving the problem of the local positron excess observed on the Earth. To clarify the latter, understanding the diffusion process is mandatory., Comment: Work presented in the 21st International Symposium on Very High Energy Cosmic Ray Interactions(ISVHECRI 2022) as part of the Ph. D. Thesis of Ramiro Torres-Escobedo (SJTU, Shanghai, China). Accepted for publication in SciPost Physics Proceedings (ISSN 2666-4003). 11 pages, 3 Figures. Short overview of HAWC and TeV Halos objects until 2022

The High-Altitude Water Cherenkov (HAWC) observatory is a second-generation continuously operated, wide field-of-view, TeV gamma-ray observatory. The HAWC observatory and its analysis techniques build on experience of the Milagro experiment in using ground-based water Cherenkov detectors for gamma-ray astronomy. HAWC is located on the Sierra Negra volcano in M\'exico at an elevation of 4100 meters above sea level. The completed HAWC observatory principal detector (HAWC) consists of 300 closely spaced water Cherenkov detectors, each equipped with four photomultiplier tubes to provide timing and charge information to reconstruct the extensive air shower energy and arrival direction. The HAWC observatory has been optimized to observe transient and steady emission from sources of gamma rays within an energy range from several hundred GeV to several hundred TeV. However, most of the air showers detected are initiated by cosmic rays, allowing studies of cosmic rays also to be performed. This paper describes the characteristics of the HAWC main array and its hardware., Comment: Accepted for publications in Nuclear Inst. and Methods in Physics Research, A (2023) 168253 ( https://www.sciencedirect.com/science/article/abs/pii/S0168900223002437 ); 39 pages, 14 Figures

Mg-based alloys are industrially used for structural applications, both as solid solutions alloys and as composites containing intermetallic compounds. However, a further development in terms of mechanical properties requires the investigation of underlying causalities between synthesis, processing and microstructure to adjust the mechanical and the corrosion properties, ideally down to the near atomic scale. Such fundamental scientific investigations with high resolution characterisation techniques require model materials of exceptionally high purity and strictly controlled microstructure e.g. with respect to grain size, morphology, chemical homogeneity as well as content and size of oxide inclusions. In this context, the Mg-Al-Ca system appears exceptionally challenging from a metallurgical perspective due to the high reactivity and high vapor pressures, so that conventional industrial techniques cannot be successfully deployed. Here, we demonstrate the applicability of various scientific synthesis methods from arc melting over solution growth to diffusion couples, extending to effects and parameters for thermo-mechanical processing. Suitable pathways to overcome the specific challenges of the Mg-Al-Ca system are demonstrated, as well as the persistent limitations of the current state of the art laboratory metallurgy technology., Comment: 17 pages, 7 figures

In this study, the structure and composition of the native oxide forming on the basal plane (0001) of Mg-2Al-0.1Ca is investigated by a correlative approach, combining scanning transmission electron microscopy (STEM) and atom probe tomography (APT). Atom probe specimens were prepared conventionally in a Ga focused ion beam (FIB) as well as a Xe plasma FIB in a cryogenic setup and subsequently cleaned in the atom probe to remove surface contamination before oxidation. While thermal energy input from the laser and longer atmospheric exposure time increased the measured hydrogen content in the specimen's apex region, cryo preparation revealed, that the hydrogen uptake in magnesium is independent of the employment of conventional or cryogenic FIB preparation. TEM measurements demonstrated the growth of a (111) MgO oxide layer with 3-4 nm thickness on the basal (0001) plane of the Mg atom probe specimen. APT data further revealed the formation of an aluminum-rich region between bulk Mg and the native oxide. The aluminum enrichment of up to ~20 at.% at the interface is consistent with an inward growth of the oxide scale.

The global trends of urbanization and increased personal mobility force us to rethink the way we live and use urban space. The Traffic4cast competition series tackles this problem in a data-driven way, advancing the latest methods in machine learning for modeling complex spatial systems over time. In this edition, our dynamic road graph data combine information from road maps, $10^{12}$ probe data points, and stationary vehicle detectors in three cities over the span of two years. While stationary vehicle detectors are the most accurate way to capture traffic volume, they are only available in few locations. Traffic4cast 2022 explores models that have the ability to generalize loosely related temporal vertex data on just a few nodes to predict dynamic future traffic states on the edges of the entire road graph. In the core challenge, participants are invited to predict the likelihoods of three congestion classes derived from the speed levels in the GPS data for the entire road graph in three cities 15 min into the future. We only provide vehicle count data from spatially sparse stationary vehicle detectors in these three cities as model input for this task. The data are aggregated in 15 min time bins for one hour prior to the prediction time. For the extended challenge, participants are tasked to predict the average travel times on super-segments 15 min into the future - super-segments are longer sequences of road segments in the graph. The competition results provide an important advance in the prediction of complex city-wide traffic states just from publicly available sparse vehicle data and without the need for large amounts of real-time floating vehicle data., Comment: Pre-print under review, submitted to Proceedings of Machine Learning Research

Decision trees are widely used for their low computational cost, good predictive performance, and ability to assess the importance of features. Though often used in practice for feature selection, the theoretical guarantees of these methods are not well understood. We here obtain a tight finite sample bound for the feature selection problem in linear regression using single-depth decision trees. We examine the statistical properties of these "decision stumps" for the recovery of the $s$ active features from $p$ total features, where $s \ll p$. Our analysis provides tight sample performance guarantees on high-dimensional sparse systems which align with the finite sample bound of $O(s \log p)$ as obtained by Lasso, improving upon previous bounds for both the median and optimal splitting criteria. Our results extend to the non-linear regime as well as arbitrary sub-Gaussian distributions, demonstrating that tree based methods attain strong feature selection properties under a wide variety of settings and further shedding light on the success of these methods in practice. As a byproduct of our analysis, we show that we can provably guarantee recovery even when the number of active features $s$ is unknown. We further validate our theoretical results and proof methodology using computational experiments., Comment: Accepted to AAAI 2023

We present work in progress using the Logarithmic Linear Relaxation (LLR) density of states algorithm to analyse first-order phase transitions in pure-gauge SU(N) Yang--Mills theories, focusing on N = 4 and 6. By using the LLR algorithm we aim to avoid super-critical slowing down at such transitions. Motivation for this study comes from composite dark matter models, which may feature a first-order confinement transition in the early Universe that would produce a background of gravitational waves. Improving our understanding of these phase transitions will help probe these models using observations from future gravitational-wave observatories. In addition to the confinement transition, we also analyze bulk phase transitions of the lattice theories, which feature much larger latent heat., Comment: arXiv admin note: text overlap with arXiv:2212.09199