Search

Concept Bottleneck Models (CBMs) have recently been proposed to address the 'black-box' problem of deep neural networks, by first mapping images to a human-understandable concept space and then linearly combining concepts for classification. Such models typically require first coming up with a set of concepts relevant to the task and then aligning the representations of a feature extractor to map to these concepts. However, even with powerful foundational feature extractors like CLIP, there are no guarantees that the specified concepts are detectable. In this work, we leverage recent advances in mechanistic interpretability and propose a novel CBM approach -- called Discover-then-Name-CBM (DN-CBM) -- that inverts the typical paradigm: instead of pre-selecting concepts based on the downstream classification task, we use sparse autoencoders to first discover concepts learnt by the model, and then name them and train linear probes for classification. Our concept extraction strategy is efficient, since it is agnostic to the downstream task, and uses concepts already known to the model. We perform a comprehensive evaluation across multiple datasets and CLIP architectures and show that our method yields semantically meaningful concepts, assigns appropriate names to them that make them easy to interpret, and yields performant and interpretable CBMs. Code available at https://github.com/neuroexplicit-saar/discover-then-name., Comment: 40 pages, 21 figures, 6 tables, European Conference on Computer Vision (ECCV) 2024

Parametric feature grid encodings have gained significant attention as an encoding approach for neural fields since they allow for much smaller MLPs, which significantly decreases the inference time of the models. In this work, we propose MeshFeat, a parametric feature encoding tailored to meshes, for which we adapt the idea of multi-resolution feature grids from Euclidean space. We start from the structure provided by the given vertex topology and use a mesh simplification algorithm to construct a multi-resolution feature representation directly on the mesh. The approach allows the usage of small MLPs for neural fields on meshes, and we show a significant speed-up compared to previous representations while maintaining comparable reconstruction quality for texture reconstruction and BRDF representation. Given its intrinsic coupling to the vertices, the method is particularly well-suited for representations on deforming meshes, making it a good fit for object animation., Comment: To appear at European Conference on Computer Vision (ECCV), 2024

Distributed execution of deep learning training involves a dynamic interplay between hardware accelerator architecture and device placement strategy. This is the first work to explore the co-optimization of determining the optimal architecture and device placement strategy through novel algorithms, improving the balance of computational resources, memory usage, and data distribution. Our architecture search leverages tensor and vector units, determining their quantity and dimensionality, and on-chip and off-chip memory configurations. It also determines the microbatch size and decides whether to recompute or stash activations, balancing the memory footprint of training and storage size. For each explored architecture configuration, we use an Integer Linear Program (ILP) to find the optimal schedule for executing operators on the accelerator. The ILP results then integrate with a dynamic programming solution to identify the most effective device placement strategy, combining data, pipeline, and tensor model parallelism across multiple accelerators. Our approach achieves higher throughput on large language models compared to the state-of-the-art TPUv4 and the Spotlight accelerator search framework. The entire source code of PHAZE is available at https://github.com/msr-fiddle/phaze., Comment: Accepted at the 41st International Conference on Machine Learning (ICML), 2024

Deep learning kernels exhibit predictable memory accesses and compute patterns, making GPUs' parallel architecture well-suited for their execution. Software and runtime systems for GPUs are optimized to better utilize the stream multiprocessors, on-chip cache, and off-chip high-bandwidth memory. As deep learning models and GPUs evolve, access to newer GPUs is often limited, raising questions about the performance of new model architectures on existing GPUs, existing models on new GPUs, and new model architectures on new GPUs. To address these questions, we introduce NeuSight, a framework to predict the performance of various deep learning models, for both training and inference, on unseen GPUs without requiring actual execution. The framework leverages both GPU hardware behavior and software library optimizations to estimate end-to-end performance. Previous work uses regression models that capture linear trends or multilayer perceptrons to predict the overall latency of deep learning kernels on GPUs. These approaches suffer from higher error percentages when forecasting performance on unseen models and new GPUs. Instead, NeuSight decomposes the prediction problem into smaller problems, bounding the prediction through fundamental performance laws. NeuSight decomposes a single deep learning kernel prediction into smaller working sets called tiles, which are executed independently on the GPU. Tile-granularity predictions are determined using a machine learning approach and aggregated to estimate end-to-end latency. NeuSight outperforms prior work across various deep learning workloads and the latest GPUs. It reduces the percentage error from 198% and 19.7% to 3.8% in predicting the latency of GPT3 model for training and inference on H100, compared to state-of-the-art prior works, where both GPT3 and H100 were not used to train the framework.

Recent JWST observations of the sub-Neptune GJ 1214 b suggest that it hosts a high-metallicity (>100x solar), hazy atmosphere. Emission spectra of the planet show molecular absorption features, most likely due to atmospheric H2O. In light of this new information, we conduct a thorough reevaluation of the planet's internal structure. We consider interior models with mixed H/He/H2O envelopes of varying composition, informed by atmospheric constraints from the JWST phase curve, in order to determine possible bulk compositions and internal structures. Self-consistent atmospheric models consistent with the JWST observations are used to set boundary conditions for the interior. We find that a total envelope mass fraction of at least 8.1% is required to explain the planet's mass and radius. Regardless of H2O content, the maximum H/He mass fraction of the planet is 5.8%. We find that a 1:1 ice-to-rock ratio along with 3.4-4.8% H/He is also a permissible solution. In addition, we consider a pure H2O (steam) envelope and find that such a scenario is possible, albeit with a high ice-to-rock ratio of at least 3.76:1, which may be unrealistic from a planet formation standpoint. We discuss possible formation pathways for the different internal structures that are consistent with observations. Since our results depend strongly on the atmospheric composition and haze properties, more precise observations of the planet's atmosphere would allow for further constraints on its internal structure. This type of analysis can be applied to any sub-Neptune with atmospheric constraints to better understand its interior., Comment: Accepted for publication in ApJ Letters. 13 pages, 6 figures, 2 tables

The ability of Language Models (LMs) to understand natural language makes them a powerful tool for parsing human instructions into task plans for autonomous robots. Unlike traditional planning methods that rely on domain-specific knowledge and handcrafted rules, LMs generalize from diverse data and adapt to various tasks with minimal tuning, acting as a compressed knowledge base. However, LMs in their standard form face challenges with long-horizon tasks, particularly in partially observable multi-agent settings. We propose an LM-based Long-Horizon Planner for Multi-Agent Robotics (LLaMAR), a cognitive architecture for planning that achieves state-of-the-art results in long-horizon tasks within partially observable environments. LLaMAR employs a plan-act-correct-verify framework, allowing self-correction from action execution feedback without relying on oracles or simulators. Additionally, we present MAP-THOR, a comprehensive test suite encompassing household tasks of varying complexity within the AI2-THOR environment. Experiments show that LLaMAR achieves a 30% higher success rate compared to other state-of-the-art LM-based multi-agent planners., Comment: 27 pages, 4 figures, 5 tables

Recently, S. Murthy has proposed a convergent expansion of free partition functions and superconformal indices of finite-$N$ purely adjoint gauge theories based on a Fredholm determinant expansion. This expansion has been dubbed the giant graviton expansion and takes the form of an infinite series of corrections to the $N=\infty$ result, with the $m^\text{th}$ correction being of order $e^{-mN}$. We show that this expansion can be reproduced using eigenvalue instantons in unitary matrix integrals. This perspective allows us to get the giant graviton expansion proposed by S. Murthy without the intermediate step of the Hubbard Stratonovich transformation., Comment: 12 pages, 1 figure

The standard approach for Partially Observable Markov Decision Processes (POMDPs) is to convert them to a fully observed belief-state MDP. However, the belief state depends on the system model and is therefore not viable in reinforcement learning (RL) settings. A widely used alternative is to use an agent state, which is a model-free, recursively updateable function of the observation history. Examples include frame stacking and recurrent neural networks. Since the agent state is model-free, it is used to adapt standard RL algorithms to POMDPs. However, standard RL algorithms like Q-learning learn a stationary policy. Our main thesis that we illustrate via examples is that because the agent state does not satisfy the Markov property, non-stationary agent-state based policies can outperform stationary ones. To leverage this feature, we propose PASQL (periodic agent-state based Q-learning), which is a variant of agent-state-based Q-learning that learns periodic policies. By combining ideas from periodic Markov chains and stochastic approximation, we rigorously establish that PASQL converges to a cyclic limit and characterize the approximation error of the converged periodic policy. Finally, we present a numerical experiment to highlight the salient features of PASQL and demonstrate the benefit of learning periodic policies over stationary policies.

Recent work has shown that object-centric representations can greatly help improve the accuracy of learning dynamics while also bringing interpretability. In this work, we take this idea one step further, ask the following question: "can learning disentangled representation further improve the accuracy of visual dynamics prediction in object-centric models?" While there has been some attempt to learn such disentangled representations for the case of static images \citep{nsb}, to the best of our knowledge, ours is the first work which tries to do this in a general setting for video, without making any specific assumptions about the kind of attributes that an object might have. The key building block of our architecture is the notion of a {\em block}, where several blocks together constitute an object. Each block is represented as a linear combination of a given number of learnable concept vectors, which is iteratively refined during the learning process. The blocks in our model are discovered in an unsupervised manner, by attending over object masks, in a style similar to discovery of slots \citep{slot_attention}, for learning a dense object-centric representation. We employ self-attention via transformers over the discovered blocks to predict the next state resulting in discovery of visual dynamics. We perform a series of experiments on several benchmark 2-D, and 3-D datasets demonstrating that our architecture (1) can discover semantically meaningful blocks (2) help improve accuracy of dynamics prediction compared to SOTA object-centric models (3) perform significantly better in OOD setting where the specific attribute combinations are not seen earlier during training. Our experiments highlight the importance discovery of disentangled representation for visual dynamics prediction.

Synthetic data is becoming increasingly important for accelerating the development of language models, both large and small. Despite several successful use cases, researchers also raised concerns around model collapse and drawbacks of imitating other models. This discrepancy can be attributed to the fact that synthetic data varies in quality and diversity. Effective use of synthetic data usually requires significant human effort in curating the data. We focus on using synthetic data for post-training, specifically creating data by powerful models to teach a new skill or behavior to another model, we refer to this setting as Generative Teaching. We introduce AgentInstruct, an extensible agentic framework for automatically creating large amounts of diverse and high-quality synthetic data. AgentInstruct can create both the prompts and responses, using only raw data sources like text documents and code files as seeds. We demonstrate the utility of AgentInstruct by creating a post training dataset of 25M pairs to teach language models different skills, such as text editing, creative writing, tool usage, coding, reading comprehension, etc. The dataset can be used for instruction tuning of any base model. We post-train Mistral-7b with the data. When comparing the resulting model Orca-3 to Mistral-7b-Instruct (which uses the same base model), we observe significant improvements across many benchmarks. For example, 40% improvement on AGIEval, 19% improvement on MMLU, 54% improvement on GSM8K, 38% improvement on BBH and 45% improvement on AlpacaEval. Additionally, it consistently outperforms other models such as LLAMA-8B-instruct and GPT-3.5-turbo.

We present NEBULA, the first latent 3D generative model for scalable generation of large molecular libraries around a seed compound of interest. Such libraries are crucial for scientific discovery, but it remains challenging to generate large numbers of high quality samples efficiently. 3D-voxel-based methods have recently shown great promise for generating high quality samples de novo from random noise (Pinheiro et al., 2023). However, sampling in 3D-voxel space is computationally expensive and use in library generation is prohibitively slow. Here, we instead perform neural empirical Bayes sampling (Saremi & Hyvarinen, 2019) in the learned latent space of a vector-quantized variational autoencoder. NEBULA generates large molecular libraries nearly an order of magnitude faster than existing methods without sacrificing sample quality. Moreover, NEBULA generalizes better to unseen drug-like molecules, as demonstrated on two public datasets and multiple recently released drugs. We expect the approach herein to be highly enabling for machine learning-based drug discovery. The code is available at https://github.com/prescient-design/nebula

Aditya-L1 is the first Indian solar mission placed at the first Lagrangian (L1) point to study the Sun. A fluxgate magnetometer (MAG) is one of the seven payloads and one of the three in-situ payloads onboard to measure the interplanetary magnetic field (IMF) coming from the Sun towards the Earth. At present, the Aditya-L1 spacecraft is in a halo-orbit around the L1 point and the MAG payload is ON is continuously measuring the IMF. This paper presents the first measurements of the IMF by MAG.

We show that dispersive gravitational waves, as a background spacetime, can reflect electromagnetic waves in a plasma. This reflection upshifts the frequency of the reflected wave, being larger for low-frequency incident waves. This effect takes place when the gravitational wave background propagates almost at the speed of light, allowing it to behave similar to a luminal mirror to electromagnetic plasma waves.

We study ZZ instanton corrections in the $(2,4k)$ $N=1$ minimal superstring theory with the type 0B GSO projection, which becomes the type 0B $N=1$ super-JT gravity in the $k \to \infty$ limit. Each member of the $(2,4k)$ family of theories has two phases distinguished by the sign of the Liouville bulk cosmological constant. The worldsheet method for computing the one-loop normalization constant multiplying the instanton corrections gives an ill-defined answer in both phases. We fix these divergences using insights from string field theory and find finite, unambiguous results. Each member of the $(2,4k)$ family of theories is dual to a double-scaled one-matrix integral, where the double-scaling limit can be obtained starting either from a unitary matrix integral with a leading one-cut saddle point, or from a hermitian matrix integral with a leading two-cut saddle point. The matrix integral exhibits a gap-closing transition, which is the same as the double-scaled Gross-Witten-Wadia transition when $k=1$. We also compute instanton corrections in the double-scaled matrix integral for all $k$ and in both phases, and find perfect agreement with the string theory results., Comment: 27 pages of main text + 27 pages of appendices + references

Instruction finetuning (IFT) is critical for aligning Large Language Models (LLMs) to follow instructions. While many effective IFT datasets have been introduced recently, they predominantly focus on high-resource languages like English. To better align LLMs across a broad spectrum of languages and tasks, we propose a fully synthetic, novel taxonomy (Evol) guided Multilingual, Multi-turn instruction finetuning dataset, called M2Lingual. It is constructed by first selecting a diverse set of seed examples and then utilizing the proposed Evol taxonomy to convert these seeds into complex and challenging multi-turn instructions. We demonstrate the effectiveness of M2Lingual by training LLMs of varying sizes and showcasing the enhanced performance across a diverse set of languages. We contribute the 2 step Evol taxonomy with the guided generation code: https://github.com/ServiceNow/M2Lingual, as well as the first fully synthetic, general and task-oriented, multi-turn, multilingual dataset built with Evol - M2Lingual: https://huggingface.co/datasets/ServiceNow-AI/ M2Lingual - containing 182K total IFT pairs, covering 70 languages and 17+ NLP tasks., Comment: 39 pages

ipie is a Python-based auxiliary-field quantum Monte Carlo (AFQMC) package that has undergone substantial improvements since its initial release [J. Chem. Theory Comput., 2022, 19(1): 109-121]. This paper outlines the improved modularity and new capabilities implemented in ipie. We highlight the ease of incorporating different trial and walker types and the seamless integration of ipie with external libraries. We enable distributed Hamiltonian simulations, allowing for multi-GPU simulations of large systems. This development enabled us to compute the interaction energy of a benzene dimer with 84 electrons and 1512 orbitals, which otherwise would not have fit on a single GPU. We also support GPU-accelerated multi-slater determinant trial wavefunctions [arXiv:2406.08314] to enable efficient and highly accurate simulations of large-scale systems. This allows for near-exact ground state energies of multi-reference clusters, [Cu$_2$O$_2$]$^{2+}$ and [Fe$_2$S$_2$(SCH$_3$)]$^{2-}$. We also describe implementations of free projection AFQMC, finite temperature AFQMC, AFQMC for electron-phonon systems, and automatic differentiation in AFQMC for calculating physical properties. These advancements position ipie as a leading platform for AFQMC research in quantum chemistry, facilitating more complex and ambitious computational method development and their applications., Comment: 17 pages, 13 figures

Recent data, added to the ITPA global H-mode database [1] for ASDEX-U [2] and JET-ILW, reveals that the separatrix density $n_{sep}$ has a correlation with the H20 factor (Confinement time relative to the ITPA20-IL scaling)[1]. These trends are analyzed in detail. They are not a result of proximity to the density limit. The normalized $n_{sepN} = n_{sep} / \bar{n}$ is introduced, motivated by theory ($\bar{n}$ is the average density). The trends in $n_{sepN}$ can be understood in terms of the two main mechanisms of pedestal characteristics -- MHD stability and recently developed theories of gyrokinetic transport. Careful analysis shows these mechanisms can be distinguished in the data. The most dramatic improvement in confinement time arises primarily from reductions in pedestal transport. A new definition of density peaking that includes core peaking is found to best explain H20 when advanced H-modes are included: $n_{sepN0} = n_{sep}/n(0)$, the inverse of the total density peaking from the separatrix to the axis. The highest H-factors are reached by the confluence of relatively low normalized $n_{sepN0}$ plus high Shafranov shift or poloidal beta. The importance of these two variables is also theoretically predicted from recent analysis of the gyrokinetic system, where a constraint can limit the access of ITG/TEM modes to free energy in equilibrium gradients. The Plasma Facing Component (PFC) material also shows a strong influence in the data. This is likely due to the importance of $n(0)/n_{sep}$ to attaining high H20, in conjunction with the known tendency for tungsten (W) to accumulate with density peaking and low transport. Preliminary results indicate that $n_{sepN0}$ might also be important with core ITBs.

We present an elementary uniform lower bound for the classical Kloosterman sum $S(a,b;c)$ under the condition of its non-vanishing and $(ab,c)=1$, with $c$ being an odd integer. We then apply this lower bound for Kloosterman sums to derive an explicit lower bound in the Petersson's trace formula, subject to a pertinent condition. Consequently, we achieve a modified version of a theorem by Jung and Sardari, wherein the parameters $k$ and $N$ are permitted to vary independently., Comment: 11 pages

Polarons are quasiparticles formed as a result of lattice distortions induced by charge carriers. The single-electron Holstein model captures the fundamentals of single polaron physics. We examine the power of the exponential ansatz for the polaron ground-state wavefunction in its coupled cluster, canonical transformation, and (canonically transformed) perturbative variants across the parameter space of the Holstein model. Our benchmark serves to guide future developments of polaron wavefunctions beyond the single-electron Holstein model., Comment: 10 pages, 6 figures

With companies, states, and countries targeting net-zero emissions around midcentury, there are questions about how these targets alter household welfare and finances, including distributional effects across income groups. This paper examines the distributional dimensions of technology transitions and net-zero policies with a focus on welfare impacts across household incomes. The analysis uses a model intercomparison with a range of energy-economy models using harmonized policy scenarios reaching economy-wide, net-zero CO2 emissions across the United States in 2050. We employ a novel linking approach that connects output from detailed energy system models with survey microdata on energy expenditures across income classes to provide distributional analysis of net-zero policies. Although there are differences in model structure and input assumptions, we find broad agreement in qualitative trends in policy incidence and energy burdens across income groups. Models generally agree that direct energy expenditures for many households will likely decline over time with reference and net-zero policies. However, there is variation in the extent of changes relative to current levels, energy burdens relative to reference levels, and electricity expenditures. Policy design, primarily how climate policy revenues are used, has first-order impacts on distributional outcomes. Net-zero policy costs, in both absolute and relative terms, are unevenly distributed across households, and relative increases in energy expenditures are higher for lowest-income households. However, we also find that recycled revenues from climate policies have countervailing effects when rebated on a per-capita basis, offsetting higher energy burdens and potentially even leading to net progressive outcomes.

There is always demand for integrating data into microeconomic decision making. Participatory sensing deals with how real-world data may be extracted with stakeholder participation and resolves a problem of Big Data, which is concerned with monetizing data extracted from individuals without their participation. We present how Decentralized Physical Infrastructure Networks (DePINs) extend participatory sensing. We discuss the threat models of these networks and how DePIN cryptoeconomics can advance participatory sensing.

We study the sample complexity of the classical shadows task: what is the fewest number of copies of an unknown state you need to measure to predict expected values with respect to some class of observables? Large joint measurements are likely required in order to minimize sample complexity, but previous joint measurement protocols only work when the unknown state is pure. We present the first joint measurement protocol for classical shadows whose sample complexity scales with the rank of the unknown state. In particular we prove $\mathcal O(\sqrt{rB}/\epsilon^2)$ samples suffice, where $r$ is the rank of the state, $B$ is a bound on the squared Frobenius norm of the observables, and $\epsilon$ is the target accuracy. In the low-rank regime, this is a nearly quadratic advantage over traditional approaches that use single-copy measurements. We present several intermediate results that may be of independent interest: a solution to a new formulation of classical shadows that captures functions of non-identical input states; a generalization of a ``nice'' Schur basis used for optimal qubit purification and quantum majority vote; and a measurement strategy that allows us to use local symmetries in the Schur basis to avoid intractable Weingarten calculations in the analysis.

In this work we demonstrate that accurate ground state wave functions may be constructed for polarons in a fully ab initio setting across the wide range of couplings associated with both the large and small polaron limits. We present a single general unitary transformation approach which encompasses an ab initio version of the Lee-Low-Pines theory at weak coupling and the coherent state Landau-Pekar framework at strong coupling while interpolating between these limits in general cases. We show that perturbation theory around these limits may be performed in a facile manner to assess the accuracy of the approach, as well as provide an independent route to the ab initio properties of polarons. We test these ideas on the case of LiF, where the electron-polaron is expected to be large and relatively weakly coupled, while the hole-polaron is expected to be a strongly coupled small polaron., Comment: 12 pages, 3 figures

In this work, we use neural quantum states (NQS) to describe the high-dimensional wave functions of electron-phonon coupled systems. We demonstrate that NQS can accurately and systematically learn the underlying physics of such problems through a variational Monte Carlo optimization of the energy with minimal incorporation of physical information even in highly challenging cases. We assess the ability of our approach across various lattice model examples featuring different types of couplings. The flexibility of our NQS formulation is demonstrated via application to ab initio models parametrized by density functional perturbation theory consisting of electron or hole bands coupled linearly to dispersive phonons. We compute accurate real-frequency spectral properties of electron-phonon systems via a novel formalism based on NQS. Our work establishes a general framework for exploring diverse ground state and dynamical phenomena arising in electron-phonon systems, including the non-perturbative interplay of correlated electronic and electron-phonon effects in systems ranging from simple lattice models to realistic models of materials parametrized by ab initio calculations., Comment: 17 pages, 13 figures

We report the synthesis and properties of (H,Li)$_{6}$Ru$_{2}$O$_{6}$, which is shown to be a $J_{\text{eff}}=\frac{1}{2}$ system made out of Ru$^{3+}$ moments in a honeycomb geometry. Bulk magnetization, heat capacity, nuclear magnetic resonance (NMR), and muon spin relaxation ($\mu$SR) rule out the presence of static moments or any spin glass phase down to 500 mK. All techniques suggest a crossover to a liquid-like state below about 40 K. The $^{7}$Li nuclear magnetic resonance (NMR) shift data suggest a non-zero $T$-independent spin susceptibility at low $T$. In zero field, $C_m/T$ shows $T^{-1}$ divergence which is consistent with vacancy-induced effects on low-energy excitations of the pristine Kitaev spin liquid. With field, power-law variations in the $^{7}$Li NMR spin-lattice relaxation rate 1/T$_{1}$ and magnetic heat capacity $C_{m}$ show quantitatively new scaling behaviors. A two-step entropy release in heat capacity is also observed putatively from $Z_{2}$ flux (low-$T$ step) and itinerant Majorana fermions (high-$T$ step). Based on these findings, we propose that (H,Li)$_{6}$Ru$_{2}$O$_{6}$ realizes a Kitaev spin liquid with no evidence of inherent magnetic ordering in zero field unlike $\alpha$-RuCl$_{3}$ where approximately $8$ Tesla field is required to suppress magnetic order., Comment: 5 pages, 6 figures of main text + 12 pages, 25 figures of supplementary information

We investigate the information-theoretic conditions to achieve the complete reconstruction of a diploid genome. We also analyze the standard greedy and de-Bruijn graph-based algorithms and compare the coverage depth and read length requirements with the information-theoretic lower bound. Our results show that the gap between the two is considerable because both algorithms require the double repeats in the genome to be bridged., Comment: Accepted at ISIT'24

In this work, we report, for the first time, an implementation of fermionic auxiliary-field quantum Monte Carlo (AFQMC) using matrix product state (MPS) trial wavefunctions, dubbed MPS-AFQMC. Calculating overlaps between an MPS trial and arbitrary Slater determinants up to a multiplicative error, a crucial subroutine in MPS-AFQMC, is proven to be #P-hard. Nonetheless, we tested several promising heuristics in successfully improving fermionic phaseless AFQMC energies. We also proposed a way to evaluate local energy and force bias evaluations free of matrix-product operators. This allows for larger basis set calculations without significant overhead. We showcase the utility of our approach on one- and two-dimensional hydrogen lattices, even when the MPS trial itself struggles to obtain high accuracy. Our work offers a new set of tools that can solve currently challenging electronic structure problems with future improvements., Comment: 24 pages, 20 figures

Air pollution is a significant health concern worldwide, contributing to various respiratory diseases. Advances in air quality mapping, driven by the emergence of smart cities and the proliferation of Internet-of-Things sensor devices, have led to an increase in available data, fueling momentum in air pollution forecasting. The objective of this study is to devise an integrated approach for predicting air quality using image data and subsequently assessing lung disease severity based on Air Quality Index (AQI).The aim is to implement an integrated approach by refining existing techniques to improve accuracy in predicting AQI and lung disease severity. The study aims to forecast additional atmospheric pollutants like AQI, PM10, O3, CO, SO2, NO2 in addition to PM2.5 levels. Additionally, the study aims to compare the proposed approach with existing methods to show its effectiveness. The approach used in this paper uses VGG16 model for feature extraction in images and neural network for predicting AQI.In predicting lung disease severity, Support Vector Classifier (SVC) and K-Nearest Neighbors (KNN) algorithms are utilized. The neural network model for predicting AQI achieved training accuracy of 88.54 % and testing accuracy of 87.44%,which was measured using loss function, while the KNN model used for predicting lung disease severity achieved training accuracy of 98.4% and testing accuracy of 97.5% In conclusion, the integrated approach presented in this study forecasts air quality and evaluates lung disease severity, achieving high testing accuracies of 87.44% for AQI and 97.5% for lung disease severity using neural network, KNN, and SVC models. The future scope involves implementing transfer learning and advanced deep learning modules to enhance prediction capabilities. While the current study focuses on India, the objective is to expand its scope to encompass global coverage., Comment: 11 pages

This study evaluates the precision of widely recognized quantum chemical methodologies, CCSD(T), DLPNO-CCSD(T) and localized ph-AFQMC, for determining the thermochemistry of main group elements. DLPNO-CCSD(T) and localized ph-AFQMC, which offer greater scalability compared to canonical CCSD(T), have emerged over the last decade as pivotal in producing precise benchmark chemical data. Our investigation includes closed-shell, neutral molecules, focusing on their heat of formation and atomization energy sourced from four specific small molecule datasets. Firstly, we selected molecules from the G2 and G3 datasets, noted for their reliable experimental heat of formation data. Additionally, we incorporate molecules from the W4-11 and W4-17 sets, which provide high-level theoretical reference values for atomization energy at 0 K. Our findings reveal that both DLPNO-CCSD(T) and ph-AFQMC methods are capable of achieving a root-mean-square deviation (RMSD) of less than 1 kcal/mol across the combined dataset, aligning with the threshold for chemical accuracy. Moreover, we make efforts to confine the maximum deviations within 2 kcal/mol, a degree of precision that significantly broadens the applicability of these methods in fields such as biology and materials science.

This review is focused on describing the logic by which we make predictions of exoplanetary compositions and mineralogies, and how these processes could lead to compositional diversity among rocky exoplanets. We use these predictions to determine the sensitivity of present-day and future observations to detecting compositional differences between rocky exoplanets and the four terrestrial planets. First, we review data on stellar abundances and infer how changes in composition may manifest themselves in the expected bulk compositions of rocky exoplanets (section 2). Converting this information in mass-radius relationships requires calculation of the stable mineral assemblages at a given temperature-pressure-composition (T-P-X), an exercise we describe in section 3. Should the planet be hot enough to engender partial melting of the mantle, then these liquids are likely to rise to the surface and erupt to form planetary crusts; the possible compositional and mineralogical variability of which we examine in section 4. Finally, the expected spectroscopic responses of such crusts are examined in section 5., Comment: Chapter 8 accepted for publication in the Reviews in Mineralogy and Geochemistry (RiMG) Volume 90 on "Exoplanets: Compositions, Mineralogy, and Evolution" edited by Natalie Hinkel, Keith Putirka, and Siyi Xu; 40 pages, 11 figures, and 3 equations

In this paper, we present a novel technique to search for hardware architectures of accelerators optimized for end-to-end training of deep neural networks (DNNs). Our approach addresses both single-device and distributed pipeline and tensor model parallel scenarios, latter being addressed for the first time. The search optimized accelerators for training relevant metrics such as throughput/TDP under a fixed area and power constraints. However, with the proliferation of specialized architectures and complex distributed training mechanisms, the design space exploration of hardware accelerators is very large. Prior work in this space has tried to tackle this by reducing the search space to either a single accelerator execution that too only for inference, or tuning the architecture for specific layers (e.g., convolution). Instead, we take a unique heuristic-based critical path-based approach to determine the best use of available resources (power and area) either for a set of DNN workloads or each workload individually. First, we perform local search to determine the architecture for each pipeline and tensor model stage. Specifically, the system iteratively generates architectural configurations and tunes the design using a novel heuristic-based approach that prioritizes accelerator resources and scheduling to critical operators in a machine learning workload. Second, to address the complexities of distributed training, the local search selects multiple (k) designs per stage. A global search then identifies an accelerator from the top-k sets to optimize training throughput across the stages. We evaluate this work on 11 different DNN models. Compared to a recent inference-only work Spotlight, our method converges to a design in, on average, 31x less time and offers 12x higher throughput. Moreover, designs generated using our method achieve 12% throughput improvement over TPU architecture.

Numerous decision-making tasks require estimating causal effects under interventions on different parts of a system. As practitioners consider using large language models (LLMs) to automate decisions, studying their causal reasoning capabilities becomes crucial. A recent line of work evaluates LLMs ability to retrieve commonsense causal facts, but these evaluations do not sufficiently assess how LLMs reason about interventions. Motivated by the role that interventions play in causal inference, in this paper, we conduct empirical analyses to evaluate whether LLMs can accurately update their knowledge of a data-generating process in response to an intervention. We create benchmarks that span diverse causal graphs (e.g., confounding, mediation) and variable types, and enable a study of intervention-based reasoning. These benchmarks allow us to isolate the ability of LLMs to accurately predict changes resulting from their ability to memorize facts or find other shortcuts. Our analysis on four LLMs highlights that while GPT- 4 models show promising accuracy at predicting the intervention effects, they remain sensitive to distracting factors in the prompts., Comment: 17 pages

Developing user-centred applications that address diverse user needs requires rigorous user research. This is time, effort and cost-consuming. With the recent rise of generative AI techniques based on Large Language Models (LLMs), there is a possibility that these powerful tools can be used to develop adaptive interfaces. This paper presents a novel approach to develop user personas and adaptive interface candidates for a specific domain using ChatGPT. We develop user personas and adaptive interfaces using both ChatGPT and a traditional manual process and compare these outcomes. To obtain data for the personas we collected data from 37 survey participants and 4 interviews in collaboration with a not-for-profit organisation. The comparison of ChatGPT generated content and manual content indicates promising results that encourage using LLMs in the adaptive interfaces design process.

We report results obtained and insights gained while answering the following question: how effective is it to use a simulator to establish path following control policies for an autonomous ground robot? While the quality of the simulator conditions the answer to this question, we found that for the simulation platform used herein, producing four control policies for path planning was straightforward once a digital twin of the controlled robot was available. The control policies established in simulation and subsequently demonstrated in the real world are PID control, MPC, and two neural network (NN) based controllers. Training the two NN controllers via imitation learning was accomplished expeditiously using seven simple maneuvers: follow three circles clockwise, follow the same circles counter-clockwise, and drive straight. A test randomization process that employs random micro-simulations is used to rank the ``goodness'' of the four control policies. The policy ranking noted in simulation correlates well with the ranking observed when the control policies were tested in the real world. The simulation platform used is publicly available and BSD3-released as open source; a public Docker image is available for reproducibility studies. It contains a dynamics engine, a sensor simulator, a ROS2 bridge, and a ROS2 autonomy stack the latter employed both in the simulator and the real world experiments., Comment: 8 pages, 7 figures

Relational network verification is a new approach to validating network changes. In contrast to traditional network verification, which analyzes specifications for a single network snapshot, relational network verification analyzes specifications concerning two network snapshots (e.g., pre- and post-change snapshots) and captures their similarities and differences. Relational change specifications are compact and precise because they specify the flows or paths that change between snapshots and then simply mandate that other behaviors of the network "stay the same", without enumerating them. To achieve similar guarantees, single-snapshot specifications need to enumerate all flow and path behaviors that are not expected to change, so we can check that nothing has accidentally changed. Thus, precise single-snapshot specifications are proportional to network size, which makes them impractical to generate for many real-world networks. To demonstrate the value of relational reasoning, we develop a high-level relational specification language and a tool called Rela to validate network changes. Rela first compiles input specifications and network snapshot representations to finite state transducers. It then checks compliance using decision procedures for automaton equivalence. Our experiments using data on complex changes to a global backbone (with over 10^3 routers) find that Rela specifications need fewer than 10 terms for 93% of them and it validates 80% of them within 20 minutes.

Training recommendation models pose significant challenges regarding resource utilization and performance. Prior research has proposed an approach that categorizes embeddings into popular and non-popular classes to reduce the training time for recommendation models. We observe that, even among the popular embeddings, certain embeddings undergo rapid training and exhibit minimal subsequent variation, resulting in saturation. Consequently, updates to these embeddings lack any contribution to model quality. This paper presents Slipstream, a software framework that identifies stale embeddings on the fly and skips their updates to enhance performance. This capability enables Slipstream to achieve substantial speedup, optimize CPU-GPU bandwidth usage, and eliminate unnecessary memory access. SlipStream showcases training time reductions of 2x, 2.4x, 1.2x, and 1.175x across real-world datasets and configurations, compared to Baseline XDL, Intel-optimized DRLM, FAE, and Hotline, respectively.

Learned indexes use machine learning models to learn the mappings between keys and their corresponding positions in key-value indexes. These indexes use the mapping information as training data. Learned indexes require frequent retrainings of their models to incorporate the changes introduced by update queries. To efficiently retrain the models, existing learned index systems often harness a linear algebraic QR factorization technique that performs matrix decomposition. This factorization approach processes all key-position pairs during each retraining, resulting in compute operations that grow linearly with the total number of keys and their lengths. Consequently, the retrainings create a severe performance bottleneck, especially for variable-length string keys, while the retrainings are crucial for maintaining high prediction accuracy and in turn, ensuring low query service latency. To address this performance problem, we develop an algorithm-hardware co-designed string-key learned index system, dubbed SIA. In designing SIA, we leverage a unique algorithmic property of the matrix decomposition-based training method. Exploiting the property, we develop a memoization-based incremental training scheme, which only requires computation over updated keys, while decomposition results of non-updated keys from previous computations can be reused. We further enhance SIA to offload a portion of this training process to an FPGA accelerator to not only relieve CPU resources for serving index queries (i.e., inference), but also accelerate the training itself. Our evaluation shows that compared to ALEX, LIPP, and SIndex, a state-of-the-art learned index systems, SIA-accelerated learned indexes offer 2.6x and 3.4x higher throughput on the two real-world benchmark suites, YCSB and Twitter cache trace, respectively., Comment: Accepted at VLDB '24; 12 pages + 2 pages (ref), 18 figures, 2 tables

Simulation can and should play a critical role in the development and testing of algorithms for autonomous agents. What might reduce its impact is the ``sim2real'' gap -- the algorithm response differs between operation in simulated versus real-world environments. This paper introduces an approach to evaluate this gap, focusing on the accuracy of sensor simulation -- specifically IMU and GPS -- in velocity estimation tasks for autonomous agents. Using a scaled autonomous vehicle, we conduct 40 real-world experiments across diverse environments then replicate the experiments in simulation with five distinct sensor noise models. We note that direct comparison of raw simulation and real sensor data fails to quantify the sim2real gap for robotics applications. We demonstrate that by using a state of the art state-estimation package as a ``judge'', and by evaluating the performance of this state-estimator in both real and simulated scenarios, we can isolate the sim2real discrepancies stemming from sensor simulations alone. The dataset generated is open-source and publicly available for unfettered use.

This paper explores the residual based a posteriori error estimations for the generalized Burgers-Huxley equation (GBHE) featuring weakly singular kernels. Initially, we present a reliable and efficient error estimator for both the stationary GBHE and the semi-discrete GBHE with memory, utilizing the discontinuous Galerkin finite element method (DGFEM) in spatial dimensions. Additionally, employing backward Euler and Crank Nicolson discretization in the temporal domain and DGFEM in spatial dimensions, we introduce an estimator for the fully discrete GBHE, taking into account the influence of past history. The paper also establishes optimal $L^2$ error estimates for both the stationary GBHE and GBHE. Ultimately, we validate the effectiveness of the proposed error estimator through numerical results, demonstrating its efficacy in an adaptive refinement strategy.

Modern transformer-based Large Language Models (LLMs) are constructed with a series of decoder blocks. Each block comprises three key components: (1) QKV generation, (2) multi-head attention, and (3) feed-forward networks. In batched processing, QKV generation and feed-forward networks involve compute-intensive matrix-matrix multiplications (GEMM), while multi-head attention requires bandwidth-heavy matrix-vector multiplications (GEMV). Machine learning accelerators like TPUs or NPUs are proficient in handling GEMM but are less efficient for GEMV computations. Conversely, Processing-in-Memory (PIM) technology is tailored for efficient GEMV computation, while it lacks the computational power to handle GEMM effectively. Inspired by this insight, we propose NeuPIMs, a heterogeneous acceleration system that jointly exploits a conventional GEMM-focused NPU and GEMV-optimized PIM devices. The main challenge in efficiently integrating NPU and PIM lies in enabling concurrent operations on both platforms, each addressing a specific kernel type. First, existing PIMs typically operate in a "blocked" mode, allowing only either NPU or PIM to be active at any given time. Second, the inherent dependencies between GEMM and GEMV in LLMs restrict their parallel processing. To tackle these challenges, NeuPIMs is equipped with dual row buffers in each bank, facilitating the simultaneous management of memory read/write operations and PIM commands. Further, NeuPIMs employs a runtime sub-batch interleaving technique to maximize concurrent execution, leveraging batch parallelism to allow two independent sub-batches to be pipelined within a single NeuPIMs device. Our evaluation demonstrates that compared to GPU-only, NPU-only, and a na\"ive NPU+PIM integrated acceleration approaches, NeuPIMs achieves 3$\times$, 2.4$\times$ and 1.6$\times$ throughput improvement, respectively., Comment: 16 pages, 15 figures

Much attention has been devoted to understanding the microscopic pathways of phase transition between two equilibrium condensed phases (such as liquids and solids). However, the microscopic pathways between non-equilibrium, non-diffusive amorphous (glassy) phases still remain poorly understood. In this work, we have employed computer simulations, persistence homology (a tool rooted in topological data analysis), and machine learning to probe the microscopic pathway of pressure-induced non-equilibrium transition between the low- and high-density amorphous (LDA and HDA, respectively) ice phases of TIP4P/2005 and ST2 water models. Using persistence homology and machine learning, we introduced a new order parameter that unambiguously identifies the LDA and HDA-like local environments. The system transitions continuously and collectively in the order parameter space via a pre-ordered intermediate phase during the compression of the LDA phase. The local order parameter susceptibilities show a maximum near the transition pressure ($P^*$) -- suggesting maximum structural heterogeneities near $P^*$. The HDA-like clusters are structurally ramified and spatially delocalized inside the LDA phase near the transition pressure. We have further investigated the (geometrical) structures and topologies of the LDA and HDA ices formed via different protocols and also studied the dependence of the microscopic pathway of phase transition in the order parameter space on the protocol followed to prepare the initial LDA phase. Finally, the method adopted here to study the microscopic pathways of transition is not restricted to the system under consideration and provides a robust way of probing phase transition pathways involving any two condensed phases at both equilibrium and out-of-equilibrium conditions.

One of the most important yet onerous tasks in the academic peer-reviewing process is composing meta-reviews, which involves understanding the core contributions, strengths, and weaknesses of a scholarly manuscript based on peer-review narratives from multiple experts and then summarizing those multiple experts' perspectives into a concise holistic overview. Given the latest major developments in generative AI, especially Large Language Models (LLMs), it is very compelling to rigorously study the utility of LLMs in generating such meta-reviews in an academic peer-review setting. In this paper, we perform a case study with three popular LLMs, i.e., GPT-3.5, LLaMA2, and PaLM2, to automatically generate meta-reviews by prompting them with different types/levels of prompts based on the recently proposed TELeR taxonomy. Finally, we perform a detailed qualitative study of the meta-reviews generated by the LLMs and summarize our findings and recommendations for prompting LLMs for this complex task.

Reinforcement learning-based methods for constructing solutions to combinatorial optimization problems are rapidly approaching the performance of human-designed algorithms. To further narrow the gap, learning-based approaches must efficiently explore the solution space during the search process. Recent approaches artificially increase exploration by enforcing diverse solution generation through handcrafted rules, however, these rules can impair solution quality and are difficult to design for more complex problems. In this paper, we introduce PolyNet, an approach for improving exploration of the solution space by learning complementary solution strategies. In contrast to other works, PolyNet uses only a single-decoder and a training schema that does not enforce diverse solution generation through handcrafted rules. We evaluate PolyNet on four combinatorial optimization problems and observe that the implicit diversity mechanism allows PolyNet to find better solutions than approaches the explicitly enforce diverse solution generation.

Text-to-image (TTI) diffusion models have demonstrated impressive results in generating high-resolution images of complex and imaginative scenes. Recent approaches have further extended these methods with personalization techniques that allow them to integrate user-illustrated concepts (e.g., the user him/herself) using a few sample image illustrations. However, the ability to generate images with multiple interacting concepts, such as human subjects, as well as concepts that may be entangled in one, or across multiple, image illustrations remains illusive. In this work, we propose a concept-driven TTI personalization framework that addresses these core challenges. We build on existing works that learn custom tokens for user-illustrated concepts, allowing those to interact with existing text tokens in the TTI model. However, importantly, to disentangle and better learn the concepts in question, we jointly learn (latent) segmentation masks that disentangle these concepts in user-provided image illustrations. We do so by introducing an Expectation Maximization (EM)-like optimization procedure where we alternate between learning the custom tokens and estimating (latent) masks encompassing corresponding concepts in user-supplied images. We obtain these masks based on cross-attention, from within the U-Net parameterized latent diffusion model and subsequent DenseCRF optimization. We illustrate that such joint alternating refinement leads to the learning of better tokens for concepts and, as a by-product, latent masks. We illustrate the benefits of the proposed approach qualitatively and quantitatively with several examples and use cases that can combine three or more entangled concepts., Comment: 11 Figures, 14 Pages, 2 tables

We consider the problem of designing a control policy for an infinite-horizon discounted cost Markov decision process $\mathcal{M}$ when we only have access to an approximate model $\hat{\mathcal{M}}$. How well does an optimal policy $\hat{\pi}^{\star}$ of the approximate model perform when used in the original model $\mathcal{M}$? We answer this question by bounding a weighted norm of the difference between the value function of $\hat{\pi}^\star $ when used in $\mathcal{M}$ and the optimal value function of $\mathcal{M}$. We then extend our results and obtain potentially tighter upper bounds by considering affine transformations of the per-step cost. We further provide upper bounds that explicitly depend on the weighted distance between cost functions and weighted distance between transition kernels of the original and approximate models. We present examples to illustrate our results.

Using JWST observations of a primary transit and two secondary eclipses for GJ 1214b, we determine an eccentricity that is more precise than a decade of HARPS data, which enables us to measure the stellar density to 2.62%. Coupled with a prior on the stellar mass from a dynamically calibrated K-$M_*$ relation, we determine $R_*$ to 1.13% -- 3 times more precise than any other published analysis of this system. Then, using the bolometric flux from an spectral energy distribution model, we determine $T_{\rm eff}$ to 1.39% -- 40% more precise than systematic floors from spectroscopy. Within the global model, these also improve the planetary radius and insolation. This is a proof of concept for a new method to determine accurate $R_*$ and $T_{\rm eff}$ to a precision currently achieved for only a small number of low-mass stars. By applying our method to all high signal-to-noise ratio planetary transits and occultations, we can expand the sample of precisely measured stars without assuming tidal circularization and calibrate new relations to improve our understanding of all low-mass stars., Comment: Published in ApJL, 13 pages, 4 figures, 4 tables. Updated to match proofs