Search

Young star-forming regions in massive environments are ideal test beds to study the influence of surroundings on the evolution of disks around low-mass stars. We explore two distant young clusters, IC 1848-East and West located in the massive W5 complex. These clusters are unique due to their similar (distance, age, and extinction) yet distinct (stellar density and FUV radiation fields) physical properties. We use deep multi-band photometry in optical, near-IR, and mid-IR wavelengths complete down to the substellar limit in at least five bands. We trace the spectral energy distribution of the sources to identify the young pre-main sequence members in the region and derive their physical parameters. The disk fraction for the East and West clusters down to 0.1 M$_\odot$ was found to be $\sim$27$\pm$2% (N$_{disk}$=184, N$_{diskless}$=492) and $\sim$17$\pm$1% (N$_{disk}$=173, N$_{diskless}$=814), respectively. While no spatial variation in the disk fraction is observed, these values are lower than those in other nearby young clusters. Investigating the cause of this decrease, we find a correlation with the intense feedback from massive stars throughout the cluster area. We also identified the disk sources undergoing accretion and observed the mass accretion rates to exhibit a positive linear relationship with the stellar host mass and an inverse relationship with stellar age. Our findings suggest that the environment significantly influences the dissipation of disks in both clusters. These distant clusters, characterized by their unique attributes, can serve as templates for future studies in outer galaxy regions, offering insights into the influence of feedback mechanisms on star and planetary formation., Comment: Accepted for publication in Monthly Notices of the Royal Astronomical Society (MNRAS). 18 pages, 11 figures, 1 table

Each request in LLM inference goes through two phases: compute-bound prefill and memory-bandwidth-bound decode. To improve GPU utilization, recent systems use hybrid batching that combines the prefill and decode phases of different requests into the same batch. Hybrid batching works well for linear operations as it amortizes the cost of loading model weights from HBM. However, attention computation in hybrid batches remains inefficient because existing attention kernels are optimized for either prefill or decode. In this paper, we present POD-Attention -- the first GPU kernel that efficiently computes attention for hybrid batches. POD-Attention aims to maximize the utilization of both compute and memory bandwidth by carefully allocating the GPU's resources such that prefill and decode operations happen concurrently on the same multiprocessor. We integrate POD-Attention in a state-of-the-art LLM inference scheduler Sarathi-Serve. POD-Attention speeds up attention computation by up to 75% (mean 28%) and increases LLM serving throughput by up to 22% in offline inference. In online inference, POD-Attention enables lower time-to-first-token (TTFT), time-between-tokens (TBT), and request execution latency versus Sarathi-Serve.

A primary goal of exoplanet science is to measure the atmospheric composition of gas giants in order to infer their formation and migration histories. Common diagnostics for planet formation are the atmospheric metallicity ([M/H]) and the carbon-to-oxygen (C/O) ratio as measured through transit or emission spectroscopy. The C/O ratio in particular can be used to approximately place a planet's initial formation radius from the stellar host, but a given C/O ratio may not be unique to formation location. This degeneracy can be broken by combining measurements of both the C/O ratio and the atmospheric refractory-to-volatile ratio. We report the measurement of both quantities for the atmosphere of the canonical ultra hot Jupiter WASP-121 b using the high resolution (R=45,000) IGRINS instrument on Gemini South. Probing the planet's direct thermal emission in both pre- and post-secondary eclipse orbital phases, we infer that WASP-121 b has a significantly super-stellar C/O ratio of 0.70$^{+0.07}_{-0.10}$ and a moderately super-stellar refractory-to-volatile ratio at 3.83$^{+3.62}_{-1.67} \times$ stellar. This combination is most consistent with formation between the soot line and H$_2$O snow line, but we cannot rule out formation between the H$_2$O and CO snow lines or beyond the CO snow line. We also measure velocity offsets between H$_2$O, CO, and OH, potentially an effect of chemical inhomogeneity on the planet day side. This study highlights the ability to measure both C/O and refractory-to-volatile ratios via high resolution spectroscopy in the near-infrared H and K bands., Comment: 21 pages, 15 figures, accepted to AJ

The chemical abundances of gas-giant exoplanet atmospheres hold clues to the formation and evolution pathways that sculpt the exoplanet population. Recent ground-based high-resolution spectroscopic observations of the non-transiting hot Jupiter $\tau$ Bo\"otis b from different instruments have resulted in a tension on the presence of water vapour in the planet's atmosphere, which impact the planet's inferred C/O and metallicity. To investigate this, we revisit the archival CRIRES observations of the planet's dayside in the wavelength range 2.28 to 2.33 $\mu$m. We reanalyse them using the latest methods for correcting stellar and telluric systematics, and free-chemistry Bayesian atmospheric retrieval. We find that a spurious detection of CH$_{4}$ can arise from inadequate telluric correction. We confirm the detection of CO and constrain its abundance to be near solar $\log_{10}(\mathrm{CO})$ = -3.44$^{+1.63}_{-0.85}$ VMR. We find a marginal evidence for H$_{2}$O with $\log_{10}(\mathrm{H_{2}O})$ = -5.13$^{+1.22}_{-6.37}$ VMR. This translates to super solar C/O (0.95$^{+0.06}_{-0.31}$), marginally sub-solar metallicity (-0.21 $^{+1.66}_{-0.87}$). Due to the relatively large uncertainty on H$_{2}$O abundance, we cannot confidently resolve the tension on the presence of H$_{2}$O and the super-solar atmospheric metallicity of $\tau$ Bo\"otis b. We recommend further observations of $\tau$ Bo\"otis b in the wavelength ranges simultaneously covering CO and $\mathrm{H_{2}O}$ to confirm the peculiar case of the planet's super-solar C/O and metallicity., Comment: 16 pages, 13 figures; Accepted for publication in MNRAS

We present a transmission spectrum of the misaligned hot Jupiter WASP-15b from 2.8--5.2 microns observed with JWST's NIRSpec/G395H grating. Our high signal to noise data, which has negligible red noise, reveals significant absorption by H$_2$O ($4.2\sigma$) and CO$_2$ ($8.9\sigma$). From independent data reduction and atmospheric retrieval approaches, we infer that WASP-15b's atmospheric metallicity is super-solar ($\gtrsim 15\times$ solar) and its C/O is consistent with solar, that together imply planetesimal accretion. Our GCM simulations for WASP-15b suggest that the C/O we measure at the limb is likely representative of the entire photosphere due to the mostly uniform spatial distribution of H$_2$O, CO$_2$ and CO. We additionally see evidence for absorption by SO$_2$ and absorption at 4.9$\mu$m, for which the current leading candidate is OCS, albeit with several caveats. If confirmed, this would be the first detection of OCS in an exoplanet atmosphere and point towards complex photochemistry of sulphur-bearing species in the upper atmosphere. These are the first observations from the BOWIE-ALIGN survey which is using JWST's NIRSpec/G395H instrument to compare the atmospheric compositions of aligned/low-obliquity and misaligned/high-obliquity hot Jupiters around F stars above the Kraft break. The goal of our survey is to determine whether the atmospheric composition differs across two populations of planets that have likely undergone different migration histories (disc versus disc-free) as evidenced by their obliquities (aligned versus misaligned)., Comment: 24 pages, 23 figures, 6 tables. Submitted to MNRAS

As virtualization of Radio Access Networks (RAN) gains momentum, understanding the impact of hardware and software disaggregation on resiliency becomes critical to meet the high availability requirements of mobile networks. Our paper presents an analytical model, using continuous time Markov chains, to study the impact of virtualization and disaggregation on RAN availability. Our evaluation, assuming typical parameter value ranges for failure and recovery rates, points to containerized platform reliability as a constraint on vRAN availability. We also find that with active-passive replication, increasing hardware replication factor beyond 2 may not bring any benefits unless failover times are reduced. We also compare the reliability of centralized and distributed virtualized central units., Comment: 7 pages, To appear at IEEE Globecom 2024

The proliferation of Internet of Things (IoT) devices has raised significant concerns regarding their security vulnerabilities. This paper explores the security risks associated with smart light systems, focusing on covert communication channels. Drawing upon previous re-search highlighting vulnerabilities in communication protocols and en-cryption flaws, the study investigates the potential for exploiting smart light systems for covert data transmission. Specifically, the paper repli-cates and analyzes an attack method introduced by Ronen and Shamir, which utilizes the Philips Hue White lighting system to create a covert channel through visible light communication (VLC). Experimental re-sults demonstrate the feasibility of transmitting data covertly through subtle variations in brightness levels, leveraging the inherent functional-ity of smart light bulbs. Despite limit. ations imposed by device constraints and communication protocols, the study underscores the need for heightened awareness and security measures in IoT environment. Ultimately, the findings emphasize the importance of implementing robust security practices and exercising caution when deploying networked IoT devices in sensitive environment., Comment: 9 pages, 2 figures , submitted in BIOM'24(COMIT'24)

Advances in spatially-resolved transcriptomics (SRT) technologies have propelled the development of new computational analysis methods to unlock biological insights. As the cost of generating these data decreases, these technologies provide an exciting opportunity to create large-scale atlases that integrate SRT data across multiple tissues, individuals, species, or phenotypes to perform population-level analyses. Here, we describe unique challenges of varying spatial resolutions in SRT data, as well as highlight the opportunities for standardized preprocessing methods along with computational algorithms amenable to atlas-scale datasets leading to improved sensitivity and reproducibility in the future., Comment: 16 pages, 2 figures

The V1298 Tau system (20-30Myr), is a benchmark young multi-planet system that provides the opportunity to perform comparative exoplanetology between planets orbiting the same star right after their formation. We present the first atmospheric comparison between two planets in the same transiting system: V1298 Tau b and V1298 Tau c. We derive constraints on the mass of planet b and c (<20M$_\oplus$ at 3$\sigma$ confidence level and $17_{-6}^{+13} M_{\oplus}$ respectively) and atmospheric metallicity (logZ/Z$_\odot$=-2.04$_{-0.59}^{0.69}$, -0.16$_{-0.94}^{1.15}$ respectively) from atmospheric retrievals. The V1298 Tau planets, are likely to be similar in terms of mass at the current age, implying that both planets are potential sub-Neptune/super-Earth progenitors. However, planet c is expected to lose a higher fraction of its mass compared to planet b given its close proximity to the host star. Alternatively, the observed spectrum of planet c can be explained by atmospheric hazes, which is in contrast to planet b where efficient haze formation can be ruled out. Higher haze formation efficiency in planet c could be due to differences in atmospheric composition, temperature and higher UV flux incident compared to planet b., Comment: Accepted for publication in Astronomy and Astrophysics

Large language models (LLMs) demonstrated transformative capabilities in many applications that require automatically generating responses based on human instruction. However, the major challenge for building LLMs, particularly in Indic languages, is the availability of high-quality data for building foundation LLMs. In this paper, we are proposing a large pre-train dataset in Hindi useful for the Indic language Hindi. We have collected the data span across several domains including major dialects in Hindi. The dataset contains 1.28 billion Hindi tokens. We have explained our pipeline including data collection, pre-processing, and availability for LLM pre-training. The proposed approach can be easily extended to other Indic and low-resource languages and will be available freely for LLM pre-training and LLM research purposes., Comment: Accepted as a book chapter in the book Title "APPLIED SPEECH AND TEXT PROCESSING FOR LOW RESOURCE LANGUAGES"

A primary objective of exoplanet atmosphere characterisation is to learn about planet formation and evolution, however, this is challenged by degeneracies. To determine whether differences in atmospheric composition can be reliably traced to differences in evolution, we are undertaking a transmission spectroscopy survey with JWST to compare the compositions of a sample of hot Jupiters that have different orbital alignments around F stars above the Kraft break. Under the assumption that aligned planets migrate through the inner disc, while misaligned planets migrate after disc dispersal, the act of migrating through the inner disc should cause a measurable difference in the C/O between aligned and misaligned planets. We expect the amplitude and sign of this difference to depend on the amount of planetesimal accretion and whether silicates accreted from the inner disc release their oxygen. Here, we identify all known exoplanets that are suitable for testing this hypothesis, describe our JWST survey, and use noise simulations and atmospheric retrievals to estimate our survey's sensitivity. With the selected sample of four aligned and four misaligned hot Jupiters, we will be sensitive to the predicted differences in C/O between aligned and misaligned hot Jupiters for a wide range of model scenarios., Comment: 14 pages, 8 figures, accepted to RASTI on 8th October 2024

Hot Jupiters present a unique opportunity for measuring how planet formation history shapes present-day atmospheric composition. However, due to the myriad pathways influencing composition, a well-constructed sample of planets is needed to determine whether formation history can be accurately traced back from atmospheric composition. To this end, the BOWIE-ALIGN survey will compare the compositions of 8 hot Jupiters around F stars, 4 with orbits aligned with the stellar rotation axis and 4 misaligned. Using the alignment as an indicator for planets that underwent disc migration or high-eccentricity migration, one can determine whether migration history produces notable differences in composition between the two samples of planets. This paper describes the planet formation model that motivates our observing programme. Our model traces the accretion of chemical components from the gas and dust in the disc over a broad parameter space to create a full, unbiased model sample from which we can estimate the range of final atmospheric compositions. For high metallicity atmospheres (O/H > 10 times solar), the C/O ratios of aligned and misaligned planets diverge, with aligned planets having lower C/O (< 0.25) due to the accretion of oxygen-rich silicates from the inner disc. However, silicates may rain out instead of releasing their oxygen into the atmosphere. This would significantly increase the C/O of aligned planets (C/O > 0.6), inverting the trend between the aligned and misaligned planets. Nevertheless, by comparing statistically significant samples of aligned and misaligned planets, we expect atmospheric composition to constrain how planets form., Comment: 11pages 10 figures, (appendix: 6 page, 4 figures), accepted to mnras (Oct 14., 2024)

The Low Latency, Low Loss, Scalable Throughput (L4S) architecture has the potential to reduce queuing delay when it is deployed at endpoints and routers throughout the Internet. However, it is not clear how TCP Prague, a prototype scalable congestion control for L4S, behaves when L4S is not yet universally deployed. Specifically, we consider the question: in a partial L4S deployment, will a user benefit by unilaterally switching from the status quo TCP to TCP Prague? To address this question, we evaluate the performance of a TCP Prague flow when sharing an L4S or non-L4S bottleneck queue with a non-L4S flow. Our findings suggest that the L4S congestion control, TCP Prague, has less favorable throughput or fairness properties than TCP Cubic or BBR in some coexistence scenarios, which may hinder adoption., Comment: Accepted to ACM Applied Networking Research Workshop (ANRW), 2024

High resolution spectroscopy of exoplanet atmospheres provides insights into their composition and dynamics from the resolved line shape and depth of thousands of spectral lines. WASP-127 b is an extremely inflated sub-Saturn (R$_\mathrm{p}$= 1.311 R$_\mathrm{Jup}$, M$_\mathrm{p}$= 0.16 M$_\mathrm{Jup}$) with previously reported detections of H$_2$O, CO$_2$, and Na. However, the seeming absence of the primary carbon reservoir expected at WASP-127 b temperatures (T$_{eq}$ $\sim$ 1400 K) from chemical equilibrium, CO, posed a mystery. In this manuscript, we present the analysis of high resolution observations of WASP-127 b with the Immersion GRating INfrared Spectrometer (IGRINS) on Gemini South. We confirm the presence of H$_2$O (8.67 $\sigma$) and report the detection of CO (4.34 $\sigma$). Additionally, we conduct a suite of Bayesian retrieval analyses covering a hierarchy of model complexity and self-consistency. When freely fitting for the molecular gas volume mixing ratios, we obtain super-solar metal enrichment for H$_2$O abundance of log$_{10}$X$_\mathrm{H_2O}$ = --1.23$^{+0.29}_{-0.49}$ and a lower limit on the CO abundance of log$_{10}$X$_\mathrm{CO}$ $\ge$ --2.20 at 2$\sigma$ confidence. We also report a tentative evidence of photochemistry in WASP-127 b based upon the indicative depletion of H$_2$S. This is also supported by the data preferring models with photochemistry over free-chemistry and thermochemistry. The overall analysis implies a super-solar ($\sim$ 39$\times$ Solar; [M/H] = $1.59^{+0.30}_{-0.30}$) metallicity for the atmosphere of WASP-127 b and an upper limit on its atmospheric C/O ratio as $<$ 0.68., Comment: 18 pages, 15 figures, submitted to AJ, poster at Exo5 conference area-A

We present a multi-wavelength analysis of the young star cluster Berkeley 59 (Be 59) based on the $Gaia$ data and deep infrared (IR) observations with the 3.58-m Telescopio Nazionale Galileo and $Spitzer$ space telescope. The mean proper motion of the cluster is found to be $\mu$$_\alpha$cos$\delta$ $\sim$ -0.63 mas yr$^{-1}$ and $\mu$$_\delta$ $\sim$ -1.83 mas yr$^{-1}$ and the kinematic distance of the cluster, $\sim$ 1 kpc, is in agreement with previous photometric studies. Present data is the deepest available near-IR observations for the cluster so far and reached below 0.03 M$_\odot$. The mass function of the cluster region is calculated using the statistically cleaned color-magnitude diagram and is similar to the Salpeter value for the member stars above 0.4 M$_\odot$. In contrast, the slope becomes shallower ($\Gamma$ $\sim$ 0.01 $\pm$ 0.18) in the mass range 0.04 - 0.4 M$_\odot$, comparable to other nearby clusters. The spatial distribution of young brown dwarfs (BDs) and stellar candidates shows a non-homogeneous distribution. This suggests that the radiation feedback from massive stars may be a prominent factor contributing to the BD population in the cluster Be 59. We also estimated the star-to-BD ratio for the cluster, which is found to be $\sim$ 3.6. The Kolomogorov-Smirnov test shows that stellar and BD populations significantly differ, and stellar candidates are near the cluster center compared to the BDs, suggesting mass segregation in the cluster toward the substellar mass regime., Comment: 18 pages, 11 Figures, accepted for the publication in the AJ

V-based XVH$_3$ (X = Li,K) hydrides perovskites are investigated for their hydrogen storage capacity using the WIEN2K code. To verify the stability of these hydrides, first-principles investigations are employed to examine their structural, electronic and hydrogen storage properties. According to structural studies these compositions hydrides are stable and part of the cubic space group (221 Pm-3m). We have examined many aspects of these compositions throughout, using the PBE-GGA exchange correlation potential. We obtained the energy versus volume curve and found the stable phase and structural parameter of these hydrides using equation of state given by Birch-Murnaghan's. These hydrides thermodynamic stability is expressed in terms of their gravimetric hydrogen storage capacity.The goal of this study is to compute the standard enthalpy of formation and thermal desorption to ascertain the stability of these hydrides. Based on band structure and density of state plots it is found that these compositions are metallic in nature. The study presents a preliminary theoretical approach for hydrogen storage applications of thermoelectric compositions, revealing their strong thermoelectric responses and potential for green energy sources.

Using the WIEN2K code, the hydrogen storage capabilities of lithium compositions like LiXH$_3$ (X = Pd, Ag, Cd) hydrides are examined. Structural, electrical, mechanical, thermoelectric, and hydrogen storage properties of these hydrides are analyzed using first-principles simulations to verify their stability. Structural analysis of these compositions reveals that the hydrides are stable and belong to the cubic space group number (221 Pm-3m). The thermodynamic stability of these hydrides are given in terms of gravimetric hydrogen storage capacities. The purpose of the study is to calculate heating of formation and breakdown temperature to determine stability of these hydrides. The metallic nature of all compositions are confirmed by band plots and density of states. The elastic properties such as elastic constant, Pugh's ratio, bulk modulus, Poisson's ratio and anisotropy factor are calculated to check the applicability of these compositions for applications involving hydrogen storage. The present paper represents the initial theoretical approach toward the future exploration of these materials for hydrogen storage applications.

Using the WIEN2K code, the hydrogen storage capabilities of lithium-based KXH$_3$ (X = Zn, Co) hydrides perovskites are examined. To verify the stability of these hydrides, first-principles simulations are employed to examine their structural, electronic, and hydrogen storage capabilities. These compositions' structural investigation shows that the hydrides are stable and part of the cubic space group (221 Pm-3m). We have examined several aspects of these composition's features throughout, using the Perdew-Burke-Ernzerhof generalized gradient approximation. The study identifies stable phases and structural parameters of hydrides using B-E equations, assessing thermodynamic stability in terms of hydrogen storage capacities. The metallic nature of these hydrides is confirmed through band structure and density calculations using WIEN2K.

With a particular focus on Scipy's minimize function the eclipse mapping method is thoroughly researched and implemented utilizing Python and essential libraries. Many optimization techniques are used, including Sequential Least Squares Programming (SLSQP), Nelder-Mead, and Conjugate Gradient (CG). However, for the purpose of examining photometric light curves these methods seek to solve the maximum entropy equation under a chi-squared constraint. Therefore, these techniques are first evaluated on two-dimensional Gaussian data without a chi-squared restriction, and then they are used to map the accretion disc and uncover the Gaussian structure of the Cataclysmic Variable KIC 201325107. Critical analysis is performed on the code structure to find possible faults and design problems. Additionally, the analysis shows how several factors impacting computing time and image quality are included including the variance in Gaussian weighting, disc image resolution, number of data points in the light curve, and degree of constraint.

Predictive biomarkers of treatment response are lacking for metastatic clear cell renal cell carcinoma (ccRCC), a tumor type that is treated with angiogenesis inhibitors, immune checkpoint inhibitors, mTOR inhibitors and a HIF2 inhibitor. The Angioscore, an RNA-based quantification of angiogenesis, is arguably the best candidate to predict anti-angiogenic (AA) response. However, the clinical adoption of transcriptomic assays faces several challenges including standardization, time delay, and high cost. Further, ccRCC tumors are highly heterogenous, and sampling multiple areas for sequencing is impractical. Here we present a novel deep learning (DL) approach to predict the Angioscore from ubiquitous histopathology slides. To overcome the lack of interpretability, one of the biggest limitations of typical DL models, our model produces a visual vascular network which is the basis of the model's prediction. To test its reliability, we applied this model to multiple cohorts including a clinical trial dataset. Our model accurately predicts the RNA-based Angioscore on multiple independent cohorts (spearman correlations of 0.77 and 0.73). Further, the predictions help unravel meaningful biology such as association of angiogenesis with grade, stage, and driver mutation status. Finally, we find our model can predict response to AA therapy, in both a real-world cohort and the IMmotion150 clinical trial. The predictive power of our model vastly exceeds that of CD31, a marker of vasculature, and nearly rivals the performance (c-index 0.66 vs 0.67) of the ground truth RNA-based Angioscore at a fraction of the cost. By providing a robust yet interpretable prediction of the Angioscore from histopathology slides alone, our approach offers insights into angiogenesis biology and AA treatment response., Comment: 19 pages, 4 Figures

The aim of this study is to teach an algorithm how to recognize different types of music. Users will submit songs for analysis. Since the algorithm hasn't heard these songs before, it needs to figure out what makes each song unique. It does this by breaking down the songs into different parts and studying things like rhythm, melody, and tone via supervised learning because the program learns from examples that are already labelled. One important thing to consider when classifying music is its genre, which can be quite complex. To ensure accuracy, we use five different algorithms, each working independently, to analyze the songs. This helps us get a more complete understanding of each song's characteristics. Therefore, our goal is to correctly identify the genre of each submitted song. Once the analysis is done, the results are presented using a graphing tool, making it easy for users to understand and provide feedback.

The signal of dipole anisotropy in quasar number counts is studied using the CatWISE2020 catalog in various color bins. It is found that the dipole signal differs significantly in two color bins, namely, $1.1>W1-W2\ge 0.8$ and $1.4>W1-W2>1.1$. The color bin $1.4>W1-W2>1.1$ appears strongly contaminated, with possibly Galactic contributions and is unreliable for extracting the signal of cosmological dipole. The source of this contamination has not been identified and cannot be attributed to known emissions within the galaxy. Removing this contaminated color bin leads to a strong dipole signal with a direction significantly different from that obtained from full data. If we interpret this dipole as due to our local motion, the extracted velocity turns out to be $900\pm 113$ Kms$^{-1}$, which deviates from the CMB dipole velocity with approximately $4.7$ sigma significance., Comment: 6 pages, 9 figures

The inner workings of neural networks can be better understood if we can fully decipher the information encoded in neural activations. In this paper, we argue that this information is embodied by the subset of inputs that give rise to similar activations. We propose InversionView, which allows us to practically inspect this subset by sampling from a trained decoder model conditioned on activations. This helps uncover the information content of activation vectors, and facilitates understanding of the algorithms implemented by transformer models. We present four case studies where we investigate models ranging from small transformers to GPT-2. In these studies, we show that InversionView can reveal clear information contained in activations, including basic information about tokens appearing in the context, as well as more complex information, such as the count of certain tokens, their relative positions, and abstract knowledge about the subject. We also provide causally verified circuits to confirm the decoded information., Comment: NeurIPS 2024; ICML 2024 Mechanistic Interpretability Workshop oral

The degree of experimentally attainable nonlocality, as gauged by the amount of loophole-free violation of Bell inequalities, remains severely limited due to inefficient detectors. We address an experimentally motivated question: Which quantum strategies attain the maximal loophole-free nonlocality in the presence of inefficient detectors? For any Bell inequality and any specification of detection efficiencies, the optimal strategies are those that maximally violate a tilted version of the Bell inequality in ideal conditions. In the simplest scenario, we demonstrate that the quantum strategies that maximally violate the tilted versions of Clauser-Horne-Shimony-Holt inequality are unique up to local isometries. However, self-testing via the standard sum of squares decomposition method turns out to be analytically intractable since even high levels of the Navascu\'es--Pironio--Ac\'in hierarchy are insufficient to saturate the maximum quantum violation of these inequalities. Instead, we utilize a novel Jordan's lemma-based proof technique to obtain robust analytical self-testing statements for the entire family of tilted-Bell inequalities. These results allow us to unveil intriguing aspects of the effect of inefficient detectors and the complexity of characterizing the set of quantum correlations, in the simplest Bell scenario., Comment: First draft, comments are welcome

Optimizing the deployment of Large language models (LLMs) is expensive today since it requires experimentally running an application workload against an LLM implementation while exploring large configuration space formed by system knobs such as parallelization strategies, batching techniques, and scheduling policies. To address this challenge, we present Vidur - a large-scale, high-fidelity, easily-extensible simulation framework for LLM inference performance. Vidur models the performance of LLM operators using a combination of experimental profiling and predictive modeling, and evaluates the end-to-end inference performance for different workloads by estimating several metrics of interest such as latency and throughput. We validate the fidelity of Vidur on several LLMs and show that it estimates inference latency with less than 9% error across the range. Further, we present Vidur-Search, a configuration search tool that helps optimize LLM deployment. Vidur-Search uses Vidur to automatically identify the most cost-effective deployment configuration that meets application performance constraints. For example, Vidur-Search finds the best deployment configuration for LLaMA2-70B in one hour on a CPU machine, in contrast to a deployment-based exploration which would require 42K GPU hours - costing ~218K dollars. Source code for Vidur is available at https://github.com/microsoft/vidur.

Efficient management of GPU memory is essential for high throughput LLM inference. Prior systems used to reserve KV-cache memory ahead-of-time that resulted in wasted capacity due to internal fragmentation. Inspired by demand paging, vLLM proposed PagedAttention to enable dynamic memory allocation for KV-cache. This approach eliminates fragmentation and improves serving throughout. However, to be able to allocate physical memory dynamically, PagedAttention changes the layout of KV-cache from contiguous virtual memory to non-contiguous virtual memory. As a consequence, one needs to rewrite the attention kernels to support paging, and implement a memory manager in the serving framework. This results in both performance and programming overheads, as well as portability challenges in adopting state-of-the-art attention kernels. In this paper, we propose vAttention, a new approach for dynamic KV-cache memory management. In contrast to PagedAttention, vAttention stores KV-cache in contiguous virtual memory and leverages OS support for on-demand allocation of physical memory. vAttention thus enables one to use state-of-the art attention kernels out-of-the-box by adding support for dynamic allocation of physical memory without having to re-write their code. We implement vAttention in the vLLM serving stack to show that it also helps improve decode throughput by up to 1.99x over vLLM, and the end-to-end serving throughput by up to 1.22x and 1.29x, compared to using the state-of-the-art PagedAttention based kernels of FlashAttention and FlashInfer., Comment: 14 pages, 13 figures, 10 tables

Transformers trained on natural language data have been shown to learn its hierarchical structure and generalize to sentences with unseen syntactic structures without explicitly encoding any structural bias. In this work, we investigate sources of inductive bias in transformer models and their training that could cause such generalization behavior to emerge. We extensively experiment with transformer models trained on multiple synthetic datasets and with different training objectives and show that while other objectives e.g. sequence-to-sequence modeling, prefix language modeling, often failed to lead to hierarchical generalization, models trained with the language modeling objective consistently learned to generalize hierarchically. We then conduct pruning experiments to study how transformers trained with the language modeling objective encode hierarchical structure. When pruned, we find joint existence of subnetworks within the model with different generalization behaviors (subnetworks corresponding to hierarchical structure and linear order). Finally, we take a Bayesian perspective to further uncover transformers' preference for hierarchical generalization: We establish a correlation between whether transformers generalize hierarchically on a dataset and whether the simplest explanation of that dataset is provided by a hierarchical grammar compared to regular grammars exhibiting linear generalization., Comment: Code now available: https://github.com/kabirahuja2431/transformers-hg

We study the data packet transmission problem (mmDPT) in dense cell-free millimeter wave (mmWave) networks, i.e., users sending data packet requests to access points (APs) via uplinks and APs transmitting requested data packets to users via downlinks. Our objective is to minimize the average delay in the system due to APs' limited service capacity and unreliable wireless channels between APs and users. This problem can be formulated as a restless multi-armed bandits problem with fairness constraint (RMAB-F). Since finding the optimal policy for RMAB-F is intractable, existing learning algorithms are computationally expensive and not suitable for practical dynamic dense mmWave networks. In this paper, we propose a structured reinforcement learning (RL) solution for mmDPT by exploiting the inherent structure encoded in RMAB-F. To achieve this, we first design a low-complexity and provably asymptotically optimal index policy for RMAB-F. Then, we leverage this structure information to develop a structured RL algorithm called mmDPT-TS, which provably achieves an \tilde{O}(\sqrt{T}) Bayesian regret. More importantly, mmDPT-TS is computation-efficient and thus amenable to practical implementation, as it fully exploits the structure of index policy for making decisions. Extensive emulation based on data collected in realistic mmWave networks demonstrate significant gains of mmDPT-TS over existing approaches., Comment: IEEE Transactions on Wireless Communications

Using starlight polarization, we present the properties of foreground dust towards cluster NGC 7380 embedded in H{\sc ii} region Sh 2-142. Observations of starlight polarization are carried out in four filters using an imaging polarimeter equipped with a 104-cm ARIES telescope. Polarization vectors of stars are aligned along the Galactic magnetic field. Towards the east and southeast regions, the dust structure appears much denser than in other regions (inferred from extinction contours and colour composite image) and is also reflected in polarization distribution. We find that the polarization degree and extinction tend to increase with distance and indication for the presence of a dust layer at a distance of around 1.2 $kpc$. We have identified eight potential candidates exhibiting intrinsic polarization by employing three distinct criteria to distinguish between stars of intrinsic polarization and interstellar polarized stars. For interstellar polarized stars, we find that the maximum polarization degree increases with the color excess and has a strong scatter, with the mean value of 1.71$\pm$0.57$\%$. The peak wavelength spans $0.40-0.88\mu$m with the mean value of 0.56$\pm$0.07 $\mu m$, suggesting similar grain sizes in the region as the average diffuse interstellar medium. The polarization efficiency is also found to decrease with visual extinction as $P_{max}/A_{V}\propto A_{V}^{-0.61}$. Our observational results are found to be consistent with the predictions by the radiative torque alignment theory., Comment: 20 pages, 7 figures, 3 tables. Accepted for publication in AJ

Each LLM serving request goes through two phases. The first is prefill which processes the entire input prompt and produces the first output token and the second is decode which generates the rest of output tokens, one-at-a-time. Prefill iterations have high latency but saturate GPU compute due to parallel processing of the input prompt. In contrast, decode iterations have low latency but also low compute utilization because a decode iteration processes only a single token per request. This makes batching highly effective for decodes and consequently for overall throughput. However, batching multiple requests leads to an interleaving of prefill and decode iterations which makes it challenging to achieve both high throughput and low latency. We introduce an efficient LLM inference scheduler, Sarathi-Serve, to address this throughput-latency tradeoff. Sarathi-Serve introduces chunked-prefills which splits a prefill request into near equal sized chunks and creates stall-free schedules that adds new requests in a batch without pausing ongoing decodes. Stall-free scheduling unlocks the opportunity to improve throughput with large batch sizes while minimizing the effect of batching on latency. Furthermore, uniform batches in Sarathi-Serve ameliorate the imbalance between iterations resulting in minimal pipeline bubbles. Our techniques yield significant improvements in inference performance across models and hardware under tail latency constraints. For Mistral-7B on single A100 GPUs, we achieve 2.6x higher serving capacity and up to 3.7x higher serving capacity for the Yi-34B model on two A100 GPUs as compared to vLLM. When used with pipeline parallelism on Falcon-180B, Sarathi-Serve provides up to 5.6x gain in the end-to-end serving capacity. The source code for Sarathi-Serve is available at https://github.com/microsoft/sarathi-serve.

We present a multi-wavelength analysis of the star cluster NGC 2316 and its surroundings. We estimated the physical parameters of the NGC 2316 cluster, including its shape (elongated), size (Rcluster = 0.4 pc), distance (1.3 +/- 0.3 kpc), and minimum reddening (AV = 1.55 mag). We discovered two massive stars (B2.0V-B1.5V, age ~12 Myr) embedded (AV = 4 mag) within this cluster. The cluster region still forms young stars even though the most massive star was born ~12 Myr ago. We also found evidence of positive feedback from these massive stars. We identified a cold gas/dust lane extending westward from the cluster. The western end of the dust lane seems to favor low-mass star formation, whereas the cluster's end favors bit massive star formation, which seems to have started earlier than the western end. We found an elongated molecular cloud in this region, characterized by numerous filamentary structures. The morphology of the filaments, along with position-velocity (pv) maps, velocity dispersion maps, channel maps, etc., indicate a coalescence of filaments and a potential longitudinal flow of matter toward the cluster through the western end of the gas/dust lane. This entire region seems to be a Hub-filamentary system (HFS), in which the NGC 2316 cluster is probably the hub and the dark lane is the main filamentary structure. Being the gravity well of this HFS, star formation started first in the NGC 2316 region and went on to the other filamentary nodes., Comment: Accepted for publication in The Astronomical Journal, 22 pages, 13 figures, 3 tables