Search

Multi-hop reasoning (MHR) is a process in artificial intelligence and natural language processing where a system needs to make multiple inferential steps to arrive at a conclusion or answer. In the context of knowledge graphs or databases, it involves traversing multiple linked entities and relationships to understand complex queries or perform tasks requiring a deeper understanding. Multi-hop reasoning is a critical function in various applications, including question answering, knowledge base completion, and link prediction. It has garnered significant interest in artificial intelligence, machine learning, and graph analytics. This paper focuses on optimizing MHR for time efficiency on large-scale graphs, diverging from the traditional emphasis on accuracy which is an orthogonal goal. We introduce a novel parallel algorithm that harnesses domain-specific learned embeddings to efficiently identify the top K paths between vertices in a knowledge graph to find the best answers to a three-hop query. Our contributions are: (1) We present a new parallel algorithm to enhance MHR performance, scalability and efficiency. (2) We demonstrate the algorithm's superior performance on leading-edge Intel and AMD architectures through empirical results. We showcase the algorithm's practicality through a case study on identifying academic affiliations of potential Turing Award laureates in Deep Learning, highlighting its capability to handle intricate entity relationships. This demonstrates the potential of our approach to enabling high-performance MHR, useful to navigate the growing complexity of modern knowledge graphs., Comment: 11 Pages with references

High-dimensional sparse data emerge in many critical application domains such as cybersecurity, healthcare, anomaly detection, and trend analysis. To quickly extract meaningful insights from massive volumes of these multi-dimensional data, scientists employ unsupervised analysis tools based on tensor decomposition (TD) methods. However, real-world sparse tensors exhibit highly irregular shapes, data distributions, and sparsity, which pose significant challenges for making efficient use of modern parallel architectures. This study breaks the prevailing assumption that compressing sparse tensors into coarse-grained structures (i.e., tensor slices or blocks) or along a particular dimension/mode (i.e., mode-specific) is more efficient than keeping them in a fine-grained, mode-agnostic form. Our novel sparse tensor representation, Adaptive Linearized Tensor Order (ALTO), encodes tensors in a compact format that can be easily streamed from memory and is amenable to both caching and parallel execution. To demonstrate the efficacy of ALTO, we accelerate popular TD methods that compute the Canonical Polyadic Decomposition (CPD) model across a range of real-world sparse tensors. Additionally, we characterize the major execution bottlenecks of TD methods on multiple generations of the latest Intel Xeon Scalable processors, including Sapphire Rapids CPUs, and introduce dynamic adaptation heuristics to automatically select the best algorithm based on the sparse tensor characteristics. Across a diverse set of real-world data sets, ALTO outperforms the state-of-the-art approaches, achieving more than an order-of-magnitude speedup over the best mode-agnostic formats. Compared to the best mode-specific formats, which require multiple tensor copies, ALTO achieves more than 5.1x geometric mean speedup at a fraction (25%) of their storage., Comment: We extend the results of our previous ICS paper to significantly improve the parallel performance of the Canonical Polyadic Alternating Least Squares (CP-ALS) algorithm for normally distributed data and the Canonical Polyadic Alternating Poisson Regression (CP-APR) algorithm for non-negative count data

In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99%) up to at least 10M tokens, a generational leap over existing models such as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75% time savings across 10 different job categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.

Artificial intelligence, particularly the subfield of machine learning, has seen a paradigm shift towards data-driven models that learn from and adapt to data. This has resulted in unprecedented advancements in various domains such as natural language processing and computer vision, largely attributed to deep learning, a special class of machine learning models. Deep learning arguably surpasses traditional approaches by learning the relevant features from raw data through a series of computational layers. This thesis explores the theoretical foundations of deep learning by studying the relationship between the architecture of these models and the inherent structures found within the data they process. In particular, we ask What drives the efficacy of deep learning algorithms and allows them to beat the so-called curse of dimensionality-i.e. the difficulty of generally learning functions in high dimensions due to the exponentially increasing need for data points with increased dimensionality? Is it their ability to learn relevant representations of the data by exploiting their structure? How do different architectures exploit different data structures? In order to address these questions, we push forward the idea that the structure of the data can be effectively characterized by its invariances-i.e. aspects that are irrelevant for the task at hand. Our methodology takes an empirical approach to deep learning, combining experimental studies with physics-inspired toy models. These simplified models allow us to investigate and interpret the complex behaviors we observe in deep learning systems, offering insights into their inner workings, with the far-reaching goal of bridging the gap between theory and practice., Comment: PhD Thesis @ EPFL

Large networks are useful in a wide range of applications. Sometimes problem instances are composed of billions of entities. Decomposing and analyzing these structures helps us gain new insights about our surroundings. Even if the final application concerns a different problem (such as traversal, finding paths, trees, and flows), decomposing large graphs is often an important subproblem for complexity reduction or parallelization. This report is a summary of discussions that happened at Dagstuhl seminar 23331 on "Recent Trends in Graph Decomposition" and presents currently open problems and future directions in the area of (hyper)graph decomposition.

Novel low-diameter network topologies such as Slim Fly (SF) offer significant cost and power advantages over the established Fat Tree, Clos, or Dragonfly. To spearhead the adoption of low-diameter networks, we design, implement, deploy, and evaluate the first real-world SF installation. We focus on deployment, management, and operational aspects of our test cluster with 200 servers and carefully analyze performance. We demonstrate techniques for simple cabling and cabling validation as well as a novel high-performance routing architecture for InfiniBand-based low-diameter topologies. Our real-world benchmarks show SF's strong performance for many modern workloads such as deep neural network training, graph analytics, or linear algebra kernels. SF outperforms non-blocking Fat Trees in scalability while offering comparable or better performance and lower cost for large network sizes. Our work can facilitate deploying SF while the associated (open-source) routing architecture is fully portable and applicable to accelerate any low-diameter interconnect.

Deep learning algorithms demonstrate a surprising ability to learn high-dimensional tasks from limited examples. This is commonly attributed to the depth of neural networks, enabling them to build a hierarchy of abstract, low-dimensional data representations. However, how many training examples are required to learn such representations remains unknown. To quantitatively study this question, we introduce the Random Hierarchy Model: a family of synthetic tasks inspired by the hierarchical structure of language and images. The model is a classification task where each class corresponds to a group of high-level features, chosen among several equivalent groups associated with the same class. In turn, each feature corresponds to a group of sub-features chosen among several equivalent ones and so on, following a hierarchy of composition rules. We find that deep networks learn the task by developing internal representations invariant to exchanging equivalent groups. Moreover, the number of data required corresponds to the point where correlations between low-level features and classes become detectable. Overall, our results indicate how deep networks overcome the curse of dimensionality by building invariant representations, and provide an estimate of the number of data required to learn a hierarchical task., Comment: 9 pages, 8 figures

Zipf's law of abbreviation, the tendency of more frequent words to be shorter, is one of the most solid candidates for a linguistic universal, in the sense that it has the potential for being exceptionless or with a number of exceptions that is vanishingly small compared to the number of languages on Earth. Since Zipf's pioneering research, this law has been viewed as a manifestation of a universal principle of communication, i.e. the minimization of word lengths, to reduce the effort of communication. Here we revisit the concordance of written language with the law of abbreviation. Crucially, we provide wider evidence that the law holds also in speech (when word length is measured in time), in particular in 46 languages from 14 linguistic families. Agreement with the law of abbreviation provides indirect evidence of compression of languages via the theoretical argument that the law of abbreviation is a prediction of optimal coding. Motivated by the need of direct evidence of compression, we derive a simple formula for a random baseline indicating that word lengths are systematically below chance, across linguistic families and writing systems, and independently of the unit of measurement (length in characters or duration in time). Our work paves the way to measure and compare the degree of optimality of word lengths in languages., Comment: In press in Glottometrics. arXiv admin note: substantial text overlap with arXiv:2208.10384

We present PolarStar, a novel family of diameter-3 network topologies derived from the star product of low-diameter factor graphs. PolarStar gives the largest known diameter-3 network topologies for almost all radixes, thus providing the best known scalable diameter-$3$ network. Compared to current state-of-the-art diameter-$3$ networks, PolarStar achieves $1.3\times$ geometric mean increase in scale over Bundlefly, $1.9\times$ over Dragonfly, and $6.7\times$ over {3-D} HyperX. PolarStar has many other desirable properties, including a modular layout, large bisection, high resilience to link failures and a large number of feasible configurations for every radix. We give a detailed evaluation with simulations of synthetic and real-world traffic patterns and show that PolarStar exhibits comparable or better performance than current diameter-3 networks., Comment: 13 pages, published in SPAA 2024

We introduce a novel technique for the reconstruction of multimode optical fields, based on simultaneously exploiting both the generalized Glauber's $K^{th}$-order correlation function $g^{(K)}$ and a recently proposed anti-correlation function (dubbed $\theta^{(K)}$) which is resilient to Poissonian noise. We experimentally demonstrate that this method yields mode reconstructions with higher fidelity with respect to those obtained with reconstruction methods based only on $g^{(K)}$'s, even requiring less "a priori" information. The reliability and versatility of our technique make it suitable for a widespread use in real applications of optical quantum measurement, from quantum information to quantum metrology, especially when one needs to characterize ensembles of single-photon emitters in the presence of background noise (due, for example, to residual excitation laser, stray light, or unwanted fluorescence)., Comment: 11 pages, 3 figures

The syntactic structure of a sentence can be represented as a graph where vertices are words and edges indicate syntactic dependencies between them. In this setting, the distance between two syntactically linked words can be defined as the difference between their positions. Here we want to contribute to the characterization of the actual distribution of syntactic dependency distances, and unveil its relationship with short-term memory limitations. We propose a new double-exponential model in which decay in probability is allowed to change after a break-point. This transition could mirror the transition from the processing of words chunks to higher-level structures. We find that a two-regime model -- where the first regime follows either an exponential or a power-law decay -- is the most likely one in all 20 languages we considered, independently of sentence length and annotation style. Moreover, the break-point is fairly stable across languages and averages values of 4-5 words, suggesting that the amount of words that can be simultaneously processed abstracts from the specific language to a high degree. Finally, we give an account of the relation between the best estimated model and the closeness of syntactic dependencies, as measured by a recently introduced optimality score.

A central question of machine learning is how deep nets manage to learn tasks in high dimensions. An appealing hypothesis is that they achieve this feat by building a representation of the data where information irrelevant to the task is lost. For image datasets, this view is supported by the observation that after (and not before) training, the neural representation becomes less and less sensitive to diffeomorphisms acting on images as the signal propagates through the net. This loss of sensitivity correlates with performance, and surprisingly correlates with a gain of sensitivity to white noise acquired during training. These facts are unexplained, and as we demonstrate still hold when white noise is added to the images of the training set. Here, we (i) show empirically for various architectures that stability to image diffeomorphisms is achieved by both spatial and channel pooling, (ii) introduce a model scale-detection task which reproduces our empirical observations on spatial pooling and (iii) compute analitically how the sensitivity to diffeomorphisms and noise scales with depth due to spatial pooling. The scalings are found to depend on the presence of strides in the net architecture. We find that the increased sensitivity to noise is due to the perturbing noise piling up during pooling, after being rectified by ReLU units.

In this short paper, we introduce the Ridgeline model, an extension of the Roofline model [4] for distributed systems. The Roofline model targets shared memory systems, bounding the performance of a kernel based on its operational intensity, and the peak compute throughput and memory bandwidth of the execution system. In a distributed setting, with multiple communicating compute entities, the network must be taken into account to model the system behavior accurately. The Ridgeline aggregates information on compute, memory, and network limits in one 2D plot to show, in an intuitive way, which of the resources is the expected bottleneck. We show the applicability of the Ridgeline in a case study based on a data-parallel Multi-Layer Perceptron (MLP) instance., Comment: 5 pages

Zipf's law of abbreviation, namely the tendency of more frequent words to be shorter, has been viewed as a manifestation of compression, i.e. the minimization of the length of forms -- a universal principle of natural communication. Although the claim that languages are optimized has become trendy, attempts to measure the degree of optimization of languages have been rather scarce. Here we present two optimality scores that are dualy normalized, namely, they are normalized with respect to both the minimum and the random baseline. We analyze the theoretical and statistical pros and cons of these and other scores. Harnessing the best score, we quantify for the first time the degree of optimality of word lengths in languages. This indicates that languages are optimized to 62 or 67 percent on average (depending on the source) when word lengths are measured in characters, and to 65 percent on average when word lengths are measured in time. In general, spoken word durations are more optimized than written word lengths in characters. Our work paves the way to measure the degree of optimality of the vocalizations or gestures of other species, and to compare them against written, spoken, or signed human languages., Comment: On the one hand, the article has been reduced: analyses of the law of abbreviation and some of the methods have been moved to another article; appendix B has been reduced. On the other hand, various parts have been rewritten for clarity; new figures have been added to ease the understanding of the scores; new citations added. Many typos have been corrected

In this paper we present PolarFly, a diameter-2 network topology based on the Erdos-Renyi family of polarity graphs from finite geometry. This is a highly scalable low-diameter topology that asymptotically reaches the Moore bound on the number of nodes for a given network degree and diameter PolarFly achieves high Moore bound efficiency even for the moderate radixes commonly seen in current and near-future routers, reaching more than 96% of the theoretical peak. It also offers more feasible router degrees than the state-of-the-art solutions, greatly adding to the selection of scalable diameter-2 networks. PolarFly enjoys many other topological properties highly relevant in practice, such as a modular design and expandability that allow incremental growth in network size without rewiring the whole network. Our evaluation shows that PolarFly outperforms competitive networks in terms of scalability, cost and performance for various traffic patterns., Comment: In Proceedings of International Conference for High Performance Computing, Networking, Storage, and Analysis (SC) 2022

It is widely believed that the success of deep networks lies in their ability to learn a meaningful representation of the features of the data. Yet, understanding when and how this feature learning improves performance remains a challenge: for example, it is beneficial for modern architectures trained to classify images, whereas it is detrimental for fully-connected networks trained for the same task on the same data. Here we propose an explanation for this puzzle, by showing that feature learning can perform worse than lazy training (via random feature kernel or the NTK) as the former can lead to a sparser neural representation. Although sparsity is known to be essential for learning anisotropic data, it is detrimental when the target function is constant or smooth along certain directions of input space. We illustrate this phenomenon in two settings: (i) regression of Gaussian random functions on the d-dimensional unit sphere and (ii) classification of benchmark datasets of images. For (i), we compute the scaling of the generalization error with number of training points, and show that methods that do not learn features generalize better, even when the dimension of the input space is large. For (ii), we show empirically that learning features can indeed lead to sparse and thereby less smooth representations of the image predictors. This fact is plausibly responsible for deteriorating the performance, which is known to be correlated with smoothness along diffeomorphisms.

Temperature is one of the most relevant parameters for the regulation of intracellular processes. Measuring localized subcellular temperature gradients is fundamental for a deeper understanding of cell function, such as the genesis of action potentials, and cell metabolism. Here, we detect for the first time temperature variations (1{\deg}C) associated with potentiation and depletion of neuronal firing, exploiting a nanoscale thermometer based on optically detected magnetic resonance in nanodiamonds. Our results provide a tool for assessing neuronal spiking activity under physiological and pathological conditions and, conjugated with the high sensitivity of this technique (in perspective sensitive to < 0.1{\deg}C variations), pave the way to a systematic study of the generation of localized temperature gradients. Furthermore, they prompt further studies explaining in detail the physiological mechanism originating this effect., Comment: 27 pages, 5 figures, 3 tables

Tensor decomposition (TD) is an important method for extracting latent information from high-dimensional (multi-modal) sparse data. This study presents a novel framework for accelerating fundamental TD operations on massively parallel GPU architectures. In contrast to prior work, the proposed Blocked Linearized Coordinate (BLCO) format enables efficient out-of-memory computation of tensor algorithms using a unified implementation that works on a single tensor copy. Our adaptive blocking and linearization strategies not only meet the resource constraints of GPU devices, but also accelerate data indexing, eliminate control-flow and memory-access irregularities, and reduce kernel launching overhead. To address the substantial synchronization cost on GPUs, we introduce an opportunistic conflict resolution algorithm, in which threads collaborate instead of contending on memory access to discover and resolve their conflicting updates on-the-fly, without keeping any auxiliary information or storing non-zero elements in specific mode orientations. As a result, our framework delivers superior in-memory performance compared to prior state-of-the-art, and is the only framework capable of processing out-of-memory tensors. On the latest Intel and NVIDIA GPUs, BLCO achieves 2.12-2.6X geometric-mean speedup (with up to 33.35X speedup) over the state-of-the-art mixed-mode compressed sparse fiber (MM-CSF) on a range of real-world sparse tensors., Comment: Accepted to ICS 2022

We study the space of supersymmetric AdS$_5$ solutions of type IIB supergravity corresponding to the conformal manifold of the dual $\mathcal{N}=1$ conformal field theory. We show that the background geometry naturally encodes a generalised holomorphic structure, dual to the superpotential of the field theory, with the existence of the full solution following from a continuity argument. In particular, this provides a solution to the long-standing problem of finding the gravity dual of the generic $\mathcal{N}=1$ deformations of $\mathcal{N}=4$ conformal field theory. Using this formalism, we derive a new result for the Hilbert series of the deformed field theories., Comment: 5 pages

Using exceptional generalised geometry, we classify which five-dimensional ${\cal N}=2$ gauged supergravities can arise as a consistent truncation of 10-/11-dimensional supergravity. Exceptional generalised geometry turns the classification into an algebraic problem of finding subgroups $G_S \subset \mathrm{USp}(8) \subset \mathrm{E}_{6(6)}$ that preserve exactly two spinors. Moreover, the intrinsic torsion of the $G_S$ structure must contain only constant singlets under $G_S$, and these, in turn, determine the gauging of the five-dimensional theory. The resulting five-dimensional theories are strongly constrained: their scalar manifolds are necessarily symmetric spaces and only a small number of matter multiplets can be kept, which we completely enumerate. We also determine the largest reductive and compact gaugings that can arise from consistent truncations., Comment: 61 pages

This paper offers a retrospective of what we learnt from organizing the workshop *Ethical Considerations in Creative applications of Computer Vision* at CVPR 2021 conference and, prior to that, a series of workshops on *Computer Vision for Fashion, Art and Design* at ECCV 2018, ICCV 2019, and CVPR 2020. We hope this reflection will bring artists and machine learning researchers into conversation around the ethical and social dimensions of creative applications of computer vision., Comment: Neural Information Processing System 2021 workshop on Machine Learning for Creativity and Design

Some recent studies have described deep convolutional neural networks to diagnose breast cancer in mammograms with similar or even superior performance to that of human experts. One of the best techniques does two transfer learnings: the first uses a model trained on natural images to create a "patch classifier" that categorizes small subimages; the second uses the patch classifier to scan the whole mammogram and create the "single-view whole-image classifier". We propose to make a third transfer learning to obtain a "two-view classifier" to use the two mammographic views: bilateral craniocaudal and mediolateral oblique. We use EfficientNet as the basis of our model. We "end-to-end" train the entire system using CBIS-DDSM dataset. To ensure statistical robustness, we test our system twice using: (a) 5-fold cross validation; and (b) the original training/test division of the dataset. Our technique reached an AUC of 0.9344 using 5-fold cross validation (accuracy, sensitivity and specificity are 85.13% at the equal error rate point of ROC). Using the original dataset division, our technique achieved an AUC of 0.8483, as far as we know the highest reported AUC for this problem, although the subtle differences in the testing conditions of each work do not allow for an accurate comparison. The inference code and model are available at https://github.com/dpetrini/two-views-classifier, Comment: Updated to published version in IEEE Access

The state-of-the-art performance for several real-world problems is currently reached by convolutional neural networks (CNN). Such learning models exploit recent results in the field of deep learning, typically leading to highly performing, yet very large neural networks with (at least) millions of parameters. As a result, the deployment of such models is not possible when only small amounts of RAM are available, or in general within resource-limited platforms, and strategies to compress CNNs became thus of paramount importance. In this paper we propose a novel lossless storage format for CNNs based on source coding and leveraging both weight pruning and quantization. We theoretically derive the space upper bounds for the proposed structures, showing their relationship with both sparsity and quantization levels of the weight matrices. Both compression rates and excution times have been tested against reference methods for matrix compression, and an empirical evaluation of state-of-the-art quantization schemes based on weight sharing is also discussed, to assess their impact on the performance when applied to both convolutional and fully connected layers. On four benchmarks for classification and regression problems and comparing to the baseline pre-trained uncompressed network, we achieved a reduction of space occupancy up to 0.6% on fully connected layers and 5.44% on the whole network, while performing at least as competitive as the baseline.

The understanding of depletion layers is of major importance to control the optical and electronic properties of metal oxide (MO) nanocrystals (NCs). Here, we show that depletion layer engineering is the main mechanism of photodoping of MO NCs. We show that the introduction of different electronic interfaces induces a double-bending of the electronic bands and a distinct carrier density profile. We found that the light-induced depletion layer modulation and bending of the bands close to the surface of the nanocrystal is the main mechanism responsible for the storage of extra electrons after photodoping in MO NCs. We support our results by a combined experimental and theoretical approach in the case of Sn:In2O3/In2O3 core-shell NCs, in which we compare numerical simulations with empirical modeling and experiments. This allows not only to extract the main mechanism of photodoping in MO NCs but also to engineer the charge storage capability of MO NCs after photodoping. Our results are transferable to other core-multishell systems, opening up a novel direction to control the optoelectronic properties of nanoscale MOs by designing their energetic band profiles through depletion layer engineering.

The Word Movers Distance (WMD) measures the semantic dissimilarity between two text documents by computing the cost of optimally moving all words of a source/query document to the most similar words of a target document. Computing WMD between two documents is costly because it requires solving an $O(V^3log(V))$ optimization problem where $V$ is the number of unique words in the document. Fortunately, WMD can be framed as an Earth Mover's Distance (EMD) for which the algorithmic complexity can be reduced to $O(V^2)$ by adding an entropy penalty to the optimization problem and solving it using the Sinkhorn-Knopp algorithm. Additionally, the computation can be made highly parallel by adopting a batching approach, i.e., computing the WMD of a single query document against multiple target documents at once. Sinkhorn WMD is a key kernel used in many ML/NLP applications. and usually gets implemented in Python. However, a straightforward Python implementation may leave significant performance on the table even though it may internally call optimized C++ BLAS routines. We present a new sparse {P}arallel {A}lgorithm for {S}inkhorn-Knopp {W}ord-movers {D}istance to compute the semantic distance of one document to many other documents by adopting the $O(V^2)$ EMD algorithm. We algorithmically transform $O(V^2)$ dense compute-heavy EMD version into an equivalent sparse one using new fused SDDMM-SpMM (sparse selection of dense-dense matrix-, sparse-dense matrix-multiplication) kernels. We implemented and optimized this algorithm for two very different architectures -- the new Intel Programmable Integrated Unified Memory Architecture (PIUMA) and Intel Xeon CPUs. We show that we were able to reach close to peak performance on both platforms., Comment: 11 Pages. arXiv admin note: substantial text overlap with arXiv:2005.06727

Understanding why deep nets can classify data in large dimensions remains a challenge. It has been proposed that they do so by becoming stable to diffeomorphisms, yet existing empirical measurements support that it is often not the case. We revisit this question by defining a maximum-entropy distribution on diffeomorphisms, that allows to study typical diffeomorphisms of a given norm. We confirm that stability toward diffeomorphisms does not strongly correlate to performance on benchmark data sets of images. By contrast, we find that the stability toward diffeomorphisms relative to that of generic transformations $R_f$ correlates remarkably with the test error $\epsilon_t$. It is of order unity at initialization but decreases by several decades during training for state-of-the-art architectures. For CIFAR10 and 15 known architectures, we find $\epsilon_t\approx 0.2\sqrt{R_f}$, suggesting that obtaining a small $R_f$ is important to achieve good performance. We study how $R_f$ depends on the size of the training set and compare it to a simple model of invariant learning., Comment: NeurIPS 2021 Conference

We consider the problem of estimating an object's physical properties such as mass, friction, and elasticity directly from video sequences. Such a system identification problem is fundamentally ill-posed due to the loss of information during image formation. Current solutions require precise 3D labels which are labor-intensive to gather, and infeasible to create for many systems such as deformable solids or cloth. We present gradSim, a framework that overcomes the dependence on 3D supervision by leveraging differentiable multiphysics simulation and differentiable rendering to jointly model the evolution of scene dynamics and image formation. This novel combination enables backpropagation from pixels in a video sequence through to the underlying physical attributes that generated them. Moreover, our unified computation graph -- spanning from the dynamics and through the rendering process -- enables learning in challenging visuomotor control tasks, without relying on state-based (3D) supervision, while obtaining performance competitive to or better than techniques that rely on precise 3D labels., Comment: ICLR 2021. Project page (and a dynamic web version of the article): https://gradsim.github.io

SU3\_Bench is a microbenchmark developed to explore performance portability across multiple programming models/methodologies using a simple, but nontrivial, mathematical kernel. This kernel has been derived from the MILC lattice quantum chromodynamics (LQCD) code. SU3\_Bench is bandwidth bound and generates regular compute and data access patterns. Therefore, on most traditional CPU and GPU-based systems, its performance is mainly determined by the achievable memory bandwidth. Although SU3\_Bench is a simple kernel, experience says its subtleties require a certain amount of tweaking to achieve peak performance for a given programming model and hardware, making performance portability challenging. In this paper, we share some of the challenges in obtaining the peak performance for SU3\_Bench on a state-of-the-art Intel Xeon machine, due to the nuances of variable definition, the nature of compiler-provided default constructors, how memory is accessed at object creation time, and the NUMA effects on the machine. We discuss how to tackle those challenges to improve SU3\_Bench's performance by \(2\times\) compared to the original OpenMP implementation available at Github. This provides a valuable lesson for other similar kernels. Expanding on the performance portability aspects, we also show early results obtained porting SU3\_Bench to the new Intel Programmable Integrated Unified Memory Architecture (PIUMA), characterized by a more balanced flops-to-byte ratio. This paper shows that it is not the usual bandwidth or flops, rather the pipeline throughput, that determines SU3\_Bench's performance on PIUMA. Finally, we show how to improve performance on PIUMA and how that compares with the performance on Xeon, which has around one order of magnitude more flops-per-byte., Comment: 11 pages

The analysis of high-dimensional sparse data is becoming increasingly popular in many important domains. However, real-world sparse tensors are challenging to process due to their irregular shapes and data distributions. We propose the Adaptive Linearized Tensor Order (ALTO) format, a novel mode-agnostic (general) representation that keeps neighboring nonzero elements in the multi-dimensional space close to each other in memory. To generate the indexing metadata, ALTO uses an adaptive bit encoding scheme that trades off index computations for lower memory usage and more effective use of memory bandwidth. Moreover, by decoupling its sparse representation from the irregular spatial distribution of nonzero elements, ALTO eliminates the workload imbalance and greatly reduces the synchronization overhead of tensor computations. As a result, the parallel performance of ALTO-based tensor operations becomes a function of their inherent data reuse. On a gamut of tensor datasets, ALTO outperforms an oracle that selects the best state-of-the-art format for each dataset, when used in key tensor decomposition operations. Specifically, ALTO achieves a geometric mean speedup of 8X over the best mode-agnostic (coordinate and hierarchical coordinate) formats, while delivering a geometric mean compression ratio of 4.3X relative to the best mode-specific (compressed sparse fiber) formats., Comment: Accepted to ICS 2021

Deep learning algorithms are responsible for a technological revolution in a variety of tasks including image recognition or Go playing. Yet, why they work is not understood. Ultimately, they manage to classify data lying in high dimension -- a feat generically impossible due to the geometry of high dimensional space and the associated curse of dimensionality. Understanding what kind of structure, symmetry or invariance makes data such as images learnable is a fundamental challenge. Other puzzles include that (i) learning corresponds to minimizing a loss in high dimension, which is in general not convex and could well get stuck bad minima. (ii) Deep learning predicting power increases with the number of fitting parameters, even in a regime where data are perfectly fitted. In this manuscript, we review recent results elucidating (i,ii) and the perspective they offer on the (still unexplained) curse of dimensionality paradox. We base our theoretical discussion on the $(h,\alpha)$ plane where $h$ is the network width and $\alpha$ the scale of the output of the network at initialization, and provide new systematic measures of performance in that plane for MNIST and CIFAR 10. We argue that different learning regimes can be organized into a phase diagram. A line of critical points sharply delimits an under-parametrised phase from an over-parametrized one. In over-parametrized nets, learning can operate in two regimes separated by a smooth cross-over. At large initialization, it corresponds to a kernel method, whereas for small initializations features can be learnt, together with invariants in the data. We review the properties of these different phases, of the transition separating them and some open questions. Our treatment emphasizes analogies with physical systems, scaling arguments and the development of numerical observables to quantitatively test these results empirically.

We discuss consistent truncations of eleven-dimensional supergravity on a six-dimensional manifold $M$, preserving minimal $\mathcal{N}=2$ supersymmetry in five dimensions. These are based on $G_S \subseteq USp(6)$ structures for the generalised $E_{6(6)}$ tangent bundle on $M$, such that the intrinsic torsion is a constant $G_S$ singlet. We spell out the algorithm defining the full bosonic truncation ansatz and then apply this formalism to consistent truncations that contain warped AdS$_5 \times_{\rm w}M$ solutions arising from M5-branes wrapped on a Riemann surface. The generalised $U(1)$ structure associated with the $\mathcal{N}=2$ solution of Maldacena-Nu\~nez leads to five-dimensional supergravity with four vector multiplets, one hypermultiplet and $SO(3)\times U(1)\times \mathbb{R}$ gauge group. The generalised structure associated with "BBBW" solutions yields two vector multiplets, one hypermultiplet and an abelian gauging. We argue that these are the most general consistent truncations on such backgrounds., Comment: 69 pages; v3: published version

Stencils represent a class of computational patterns where an output grid point depends on a fixed shape of neighboring points in an input grid. Stencil computations are prevalent in scientific applications engaging a significant portion of supercomputing resources. Therefore, it has been always important to optimize stencil programs for the best performance. A rich body of research has focused on optimizing stencil computations on almost all parallel architectures. Stencil applications have regular dependency patterns, inherent pipeline-parallelism, and plenty of data reuse. This makes these applications a perfect match for a coarse-grained reconfigurable spatial architecture (CGRA). A CGRA consists of many simple, small processing elements (PEs) connected with an on-chip network. Each PE can be configured to execute part of a stencil computation and all PEs run in parallel; the network can also be configured so that data loaded can be passed from a PE to a neighbor PE directly and thus reused by many PEs without register spilling and memory traffic. How to efficiently map a stencil computation to a CGRA is the key to performance. In this paper, we show a few unique and generalizable ways of mapping one- and multidimensional stencil computations to a CGRA, fully exploiting the data reuse opportunities and parallelism. Our simulation experiments demonstrate that these mappings are efficient and enable the CGRA to outperform state-of-the-art GPUs., Comment: 9 Pages

Zero-shot classification is a generalization task where no instance from the target classes is seen during training. To allow for test-time transfer, each class is annotated with semantic information, commonly in the form of attributes or text descriptions. While classical zero-shot learning does not explicitly forbid using information from other datasets, the approaches that achieve the best absolute performance on image benchmarks rely on features extracted from encoders pretrained on Imagenet. This approach relies on hyper-optimized Imagenet-relevant parameters from the supervised classification setting, entangling important questions about the suitability of those parameters and how they were learned with more fundamental questions about representation learning and generalization. To remove these distractors, we propose a more challenging setting: Zero-Shot Learning from scratch (ZFS), which explicitly forbids the use of encoders fine-tuned on other datasets. Our analysis on this setting highlights the importance of local information, and compositional representations., Comment: ICML 2019 Workshop on Understanding and Improving General-ization in Deep Learning, Long Beach, California, 2019 Spotlight presentation. arXiv admin note: text overlap with arXiv:1912.12179

High performance large scale graph analytics is essential to timely analyze relationships in big data sets. Conventional processor architectures suffer from inefficient resource usage and bad scaling on graph workloads. To enable efficient and scalable graph analysis, Intel developed the Programmable Integrated Unified Memory Architecture (PIUMA). PIUMA consists of many multi-threaded cores, fine-grained memory and network accesses, a globally shared address space and powerful offload engines. This paper presents the PIUMA architecture, and provides initial performance estimations, projecting that a PIUMA node will outperform a conventional compute node by one to two orders of magnitude. Furthermore, PIUMA continues to scale across multiple nodes, which is a challenge in conventional multinode setups.

We study how neural networks compress uninformative input space in models where data lie in $d$ dimensions, but whose label only vary within a linear manifold of dimension $d_\parallel < d$. We show that for a one-hidden layer network initialized with infinitesimal weights (i.e. in the feature learning regime) trained with gradient descent, the first layer of weights evolve to become nearly insensitive to the $d_\perp=d-d_\parallel$ uninformative directions. These are effectively compressed by a factor $\lambda\sim \sqrt{p}$, where $p$ is the size of the training set. We quantify the benefit of such a compression on the test error $\epsilon$. For large initialization of the weights (the lazy training regime), no compression occurs and for regular boundaries separating labels we find that $\epsilon \sim p^{-\beta}$, with $\beta_\text{Lazy} = d / (3d-2)$. Compression improves the learning curves so that $\beta_\text{Feature} = (2d-1)/(3d-2)$ if $d_\parallel = 1$ and $\beta_\text{Feature} = (d + d_\perp/2)/(3d-2)$ if $d_\parallel > 1$. We test these predictions for a stripe model where boundaries are parallel interfaces ($d_\parallel=1$) as well as for a cylindrical boundary ($d_\parallel=2$). Next we show that compression shapes the Neural Tangent Kernel (NTK) evolution in time, so that its top eigenvectors become more informative and display a larger projection on the labels. Consequently, kernel learning with the frozen NTK at the end of training outperforms the initial NTK. We confirm these predictions both for a one-hidden layer FC network trained on the stripe model and for a 16-layers CNN trained on MNIST, for which we also find $\beta_\text{Feature}>\beta_\text{Lazy}$.

Understanding the human brain remains one of the most significant challenges of the 21st century. As theoretical studies continue to improve the description of the complex mechanisms that regulate biological processes, in parallel numerous experiments are conducted to enrich or verify these theoretical predictions and with the aim of extrapolating more accurate models. In the field of magnetometers for biological application, among the various sensors proposed for this purpose, NV centers have emerged as a promising solution due to their perfect biocompatibility and the possibility of being positioned in close proximity and even inside the cell, allowing a nanometric spatial resolution. There are still many difficulties that must be overcome in order to obtain both spatial resolution and sensitivity capable of revealing the very weak biological electromagnetic fields generated by neurons (or other cells). However, over the last few years, significant improvements have been achieved in this direction, thanks to the use of innovative techniques, which allow us to hope for an early application of these sensors for the measurement of fields such as the one generated by cardiac tissue, if not, in perspective, for the nerve fibers fields. In this review, we will analyze the new results regarding the application of NV centers and we will discuss the main challenges that currently prevent these quantum sensors from reaching their full potential., Comment: 21 pages, 13 figures

The possibility of using Nitrogen-vacancy centers in diamonds to measure nanoscale magnetic fields with unprecedented sensitivity is one of the most significant achievements of quantum sensing. Here we present an innovative experimental set-up, showing an achieved sensitivity comparable to the state of the art ODMR protocols if the sensing volume is taken into account. The apparatus allows magnetic sensing in biological samples such as individual cells, as it is characterized by a small sensing volume and full bio-compatibility. The sensitivity at different optical powers is studied to extend this technique to the intercellular scale., Comment: 6 pages, 5 figures

The Word Mover's Distance (WMD) is a metric that measures the semantic dissimilarity between two text documents by computing the cost of moving all words of a source/query document to the most similar words of a target document optimally. Computing WMD between two documents is costly because it requires solving an optimization problem that costs \(O(V^3log(V))\) where \(V\) is the number of unique words in the document. Fortunately, the WMD can be framed as the Earth Mover's Distance (EMD) (also known as the Optimal Transportation Distance) for which it has been shown that the algorithmic complexity can be reduced to \(O(V^2)\) by adding an entropy penalty to the optimization problem and a similar idea can be adapted to compute WMD efficiently. Additionally, the computation can be made highly parallel by computing WMD of a single query document against multiple target documents at once (e.g., finding whether a given tweet is similar to any other tweets happened in a day). In this paper, we present a shared-memory parallel Sinkhorn-Knopp Algorithm to compute the WMD of one document against many other documents by adopting the \(O(V^2)\) EMD algorithm. We used algorithmic transformations to change the original dense compute-heavy kernel to a sparse compute kernel and obtained \(67\times\) speedup using \(96\) cores on the state-of-the-art of Intel\textregistered{} 4-sockets Cascade Lake machine w.r.t. its sequential run. Our parallel algorithm is over \(700\times\) faster than the naive parallel python code that internally uses optimized matrix library calls., Comment: 10 pages, 1 page for reference, total 11 pages

Online and Real-time Object Tracking is an interesting workload that can be used to track objects (e.g., car, human, animal) in a series of video sequences in real-time. For simple object tracking on edge devices, the output of object tracking could be as simple as drawing a bounding box around a detected object and in some cases, the input matrices used in such computation are quite small (e.g., 4x7, 3x3, 5x5, etc). As a result, the amount of actual work is low. Therefore, a typical multi-threading based parallelization technique can not accelerate the tracking application; instead, a throughput based parallelization technique where each thread operates on independent video sequences is more rewarding. In this paper, we share our experience in parallelizing a Simple Online and Real-time Tracking (SORT) application on shared-memory multicores., Comment: 5 Pages (4 Pages main paper, 5th page for reference), Accepted for presentation in WHPC 2020 Summit which got canceled for Corona. But it will not be published in Digital Library

In this work we study locality and compositionality in the context of learning representations for Zero Shot Learning (ZSL). In order to well-isolate the importance of these properties in learned representations, we impose the additional constraint that, differently from most recent work in ZSL, no pre-training on different datasets (e.g. ImageNet) is performed. The results of our experiments show how locality, in terms of small parts of the input, and compositionality, i.e. how well can the learned representations be expressed as a function of a smaller vocabulary, are both deeply related to generalization and motivate the focus on more local-aware models in future research directions for representation learning., Comment: Published at ICLR 2020

We present a generalised geometry framework for systematically constructing consistent truncations of ten- and eleven-dimensional supergravity preserving varying fractions of supersymmetry. Truncations arise when there is a reduced structure group $G_S$ of the exceptional generalised geometry, such that the intrinsic torsion is a $G_S$-singlet. The matter content of the truncated theory follows from group-theoretical arguments, while the gauging is determined by the sub-algebra of generalised diffeomorphisms generated by the $G_S$-singlet vectors. After discussing the general ideas across different spacetime dimensions and amounts of supersymmetry, we provide detailed formulae for truncations to gauged half-maximal supergravity in five dimensions. In particular, we establish an expression for the generalised metric on the exceptional tangent bundle, which determines the scalar truncation ansatz. As applications, we show that this formalism gives a simple derivation of a new consistent truncation of type IIB supergravity on $\beta$-deformed Lunin-Maldacena geometries, yielding half-maximal supergravity coupled to two vector multiplets, and of the truncation of eleven-dimensional supergravity on Maldacena-N\'u\~nez geometries, given by $S^4$ twisted over a Riemann surface, which leads to half-maximal supergravity coupled to three vector multiplets., Comment: 60 pages; v2: minor changes, added references

With the advent of powerful, low-cost IoT systems, processing data closer to where the data originates, known as edge computing, has become an increasingly viable option. In addition to lowering the cost of networking infrastructures, edge computing reduces edge-cloud delay, which is essential for mission-critical applications. In this paper, we show the feasibility and study the performance of image classification using IoT devices. Specifically, we explore the relationships between various factors of image classification algorithms that may affect energy consumption such as dataset size, image resolution, algorithm type, algorithm phase, and device hardware. Our experiments show a strong, positive linear relationship between three predictor variables, namely model complexity, image resolution, and dataset size, with respect to energy consumption. In addition, in order to provide a means of predicting the energy consumption of an edge device performing image classification, we investigate the usage of three machine learning algorithms using the data generated from our experiments. The performance as well as the trade offs for using linear regression, Gaussian process, and random forests are discussed and validated. Our results indicate that the random forest model outperforms the two former algorithms, with an R-squared value of 0.95 and 0.79 for two different validation datasets.

We present the uplift of the GPPZ solution of the five-dimensional maximal supergravity to ten dimensions. The five dimensional solution involves two real scalar fields, with one of them encoding holographically the (norm of the complex) supersymmetric ${\mathcal N}=1$ mass deformation and the other the real part of the gaugino condensate. We embed this solution in a consistent truncation of $D=5$ maximal supergravity which involves two complex scalars dual to the complex mass deformations and the complex gaugino condensate, and a $U(1)$ gauge field dual to the $U(1)_R$ current, and uplift it to ten dimensions. The ten dimensional solution is completely explicit, with all fields given in terms of elementary functions. The metric and the axion-dilaton agree with those of a partial uplift of the GPPZ flow by Pilch and Warner. We analyze the asymptotics and the singularity structure of the ten dimensional solution. The uplifted solution is singular, but the singularity is milder than that of the five dimensional solution, and there is conformal frame in which the metric is only singular at one point of $S^5$. We compare the asymptotics of the $10d$ solution with that of the Polchinski-Strassler and Freedman-Minahan solutions, and find agreement with Freedman-Minahan and disagreement with Polchinski-Strassler. In particular, we infer that while the Polchinski-Strassler $10d$ fields satisfy the correct boundary conditions, they do not solve the field equations near the boundary., Comment: 45 pages, v3: JHEP version, Mathematica file with the IIB solution and the verification of Einstein equations attached, corrected some typos, v4: Note Added, typos corrected

We study heterotic backgrounds with non-trivial $H$-flux and non-vanishing expectation values of fermionic bilinears, often referred to as gaugino condensates. The gaugini appear in the low energy action via the gauge-invariant three-form bilinear $\Sigma_{MNP}={\rm tr}\:\bar\chi\Gamma_{MNP}\chi$. For Calabi-Yau compactifications to four dimensions, the gaugino condensate corresponds to an internal three-form $\Sigma_{mnp}$ that must be a singlet of the holonomy group. This condition does not hold anymore when an internal $H$-flux is turned on and ${\cal O}(\alpha')$ effects are included. In this paper we study flux compactifications to three and four-dimensions on $G$-structure manifolds. We derive the generic conditions for supersymmetric solutions. We use integrability conditions and Lichnerowicz type arguments to derive a set of constraints whose solution, together with supersymmetry, is sufficient for finding backgrounds with gaugino condensate., Comment: 44 pages, minor corrections, references added

We apply exceptional generalised geometry to the study of exactly marginal deformations of $\mathcal{N}=1$ SCFTs that are dual to generic AdS$_5$ flux backgrounds in type IIB or eleven-dimensional supergravity. In the gauge theory, marginal deformations are parametrised by the space of chiral primary operators of conformal dimension three, while exactly marginal deformations correspond to quotienting this space by the complexified global symmetry group. We show how the supergravity analysis gives a geometric interpretation of the gauge theory results. The marginal deformations arise from deformations of generalised structures that solve moment maps for the generalised diffeomorphism group and have the correct charge under the generalised Reeb vector, generating the R-symmetry. If this is the only symmetry of the background, all marginal deformations are exactly marginal. If the background possesses extra isometries, there are obstructions that come from fixed points of the moment maps. The exactly marginal deformations are then given by a further quotient by these extra isometries. Our analysis holds for any $\mathcal{N}=2$ AdS$_5$ flux background. Focussing on the particular case of type IIB Sasaki-Einstein backgrounds we recover the result that marginal deformations correspond to perturbing the solution by three-form flux at first order. In various explicit examples, we show that our expression for the three-form flux matches those in the literature and the obstruction conditions match the one-loop beta functions of the dual SCFT., Comment: 52 pages

We develop an exceptional generalised geometry formalism for massive type IIA supergravity. In particular, we construct a deformation of the generalised Lie derivative, which generates the type IIA gauge transformations as modified by the Romans mass. We apply this new framework to consistent Kaluza-Klein reductions preserving maximal supersymmetry. We find a generalised parallelisation of the exceptional tangent bundle on S^6, and from this reproduce the consistent truncation ansatz and embedding tensor leading to dyonically gauged ISO(7) supergravity in four dimensions. We also discuss closely related hyperboloid reductions, yielding a dyonic ISO(p,7-p) gauging. Finally, while for vanishing Romans mass we find a generalised parallelisation on S^d, d=4,3,2, leading to a maximally supersymmetric reduction with gauge group SO(d+1) (or larger), we provide evidence that an analogous reduction does not exist in the massive theory., Comment: 69 pages; v2: version published in JHEP

The problem of counting occurrences of query graphs in a large data graph, known as subgraph counting, is fundamental to several domains such as genomics and social network analysis. Many important special cases (e.g. triangle counting) have received significant attention. Color coding is a very general and powerful algorithmic technique for subgraph counting. Color coding has been shown to be effective in several applications, but scalable implementations are only known for the special case of {\em tree queries} (i.e. queries of treewidth one). In this paper we present the first efficient distributed implementation for color coding that goes beyond tree queries: our algorithm applies to any query graph of treewidth $2$. Since tree queries can be solved in time linear in the size of the data graph, our contribution is the first step into the realm of colour coding for queries that require superlinear running time in the worst case. This superlinear complexity leads to significant load balancing problems on graphs with heavy tailed degree distributions. Our algorithm structures the computation to work around high degree nodes in the data graph, and achieves very good runtime and scalability on a diverse collection of data and query graph pairs as a result. We also provide theoretical analysis of our algorithmic techniques, showing asymptotic improvements in runtime on random graphs with power law degree distributions, a popular model for real world graphs.

We analyse generic AdS flux backgrounds preserving eight supercharges in $D=4$ and $D=5$ dimensions using exceptional generalised geometry. We show that they are described by a pair of globally defined, generalised structures, identical to those that appear for flat flux backgrounds but with different integrability conditions. We give a number of explicit examples of such "exceptional Sasaki-Einstein" backgrounds in type IIB supergravity and M-theory. In particular, we give the complete analysis of the generic AdS$_5$ M-theory backgrounds. We also briefly discuss the structure of the moduli space of solutions. In all cases, one structure defines a "generalised Reeb vector" that generates a Killing symmetry of the background corresponding to the R-symmetry of the dual field theory, and in addition encodes the generic contact structures that appear in the $D=4$ M-theory and $D=5$ type IIB cases. Finally, we investigate the relation between generalised structures and quantities in the dual field theory, showing that the central charge and R-charge of BPS wrapped-brane states are both encoded by the generalised Reeb vector, as well as discussing how volume minimisation (the dual of $a$- and $\mathcal{F}$-maximisation) is encoded., Comment: 40 pages

We examine the AdS-CFT dual of arbitrary (non)supersymmetric fermionic mass deformations of N=4 SYM, and investigate how the backreaction of the RR and NS-NS two-form potentials dual to the fermion masses contribute to Coulomb-branch potential of D3 branes, which we interpret as the bulk boson mass matrix. Using representation-theory and supergravity arguments we show that the fermion masses completely determine the trace of this matrix, and that on the other hand its traceless components have to be turned on as non-normalizable modes. Our result resolves the tension between the belief that the AdS bulk dual of the trace of the boson mass matrix (which is not a chiral operator) is a stringy excitation with dimension of order $(g_s N)^{1/4}$ and the existence of non-stringy supergravity flows describing theories where this trace is nonzero, by showing that the stringy mode does not parameterize the sum of the squares of the boson masses but rather its departure from the trace of the square of the fermion mass matrix. Hence, asymptotically-AdS flows can only describe holographically theories where the sums of the squares of the bosonic and fermionic masses are equal, which is consistent with the weakly-coupled result that only such theories can have a conformal UV fixed point., Comment: 15 pages, LateX

A common way to obtain standard-model-like Lagrangians in string theory is to place D3-branes inside flux compactifications. The bosonic and fermionic masses and couplings of the resulting gauge theory are determined by the ten-dimensional metric and the fluxes, respectively, and the breaking of supersymmetry is soft. However, not any soft-supersymmetry-breaking Lagrangian can be obtained this way since the string theory equations of motion impose certain relations between the soft couplings. We show that for D3-branes in background fluxes, these relations imply that the sums of the squares of the boson and of the fermion masses are equal and that, furthermore, one- and two-loop quantum corrections do not spoil this equality. This makes the use of D3-branes for constructing computationally controllable models for physics beyond the standard model problematic., Comment: 4 pages. v2: minor changes