Search

Frameworks for writing, compiling, and optimizing deep learning (DL) models have recently enabled progress in areas like computer vision and natural language processing. Extending these frameworks to accommodate the rapidly diversifying landscape of DL models and hardware platforms presents challenging tradeoffs between expressivity, composability, and portability. We present Relay, a new compiler framework for DL. Relay's functional, statically typed intermediate representation (IR) unifies and generalizes existing DL IRs to express state-of-the-art models. The introduction of Relay's expressive IR requires careful design of domain-specific optimizations, addressed via Relay's extension mechanisms. Using these extension mechanisms, Relay supports a unified compiler that can target a variety of hardware platforms. Our evaluation demonstrates Relay's competitive performance for a broad class of models and devices (CPUs, GPUs, and emerging accelerators). Relay's design demonstrates how a unified IR can provide expressivity, composability, and portability without compromising performance.

There is an increasing need to bring machine learning to a wide diversity of hardware devices. Current frameworks rely on vendor-specific operator libraries and optimize for a narrow range of server-class GPUs. Deploying workloads to new platforms -- such as mobile phones, embedded devices, and accelerators (e.g., FPGAs, ASICs) -- requires significant manual effort. We propose TVM, a compiler that exposes graph-level and operator-level optimizations to provide performance portability to deep learning workloads across diverse hardware back-ends. TVM solves optimization challenges specific to deep learning, such as high-level operator fusion, mapping to arbitrary hardware primitives, and memory latency hiding. It also automates optimization of low-level programs to hardware characteristics by employing a novel, learning-based cost modeling method for rapid exploration of code optimizations. Experimental results show that TVM delivers performance across hardware back-ends that are competitive with state-of-the-art, hand-tuned libraries for low-power CPU, mobile GPU, and server-class GPUs. We also demonstrate TVM's ability to target new accelerator back-ends, such as the FPGA-based generic deep learning accelerator. The system is open sourced and in production use inside several major companies., Comment: Significantly improved version, add automated optimization

State of the art deep learning models have made steady progress in the fields of computer vision and natural language processing, at the expense of growing model sizes and computational complexity. Deploying these models on low power and mobile devices poses a challenge due to their limited compute capabilities and strict energy budgets. One solution that has generated significant research interest is deploying highly quantized models that operate on low precision inputs and weights less than eight bits, trading off accuracy for performance. These models have a significantly reduced memory footprint (up to 32x reduction) and can replace multiply-accumulates with bitwise operations during compute intensive convolution and fully connected layers. Most deep learning frameworks rely on highly engineered linear algebra libraries such as ATLAS or Intel's MKL to implement efficient deep learning operators. To date, none of the popular deep learning directly support low precision operators, partly due to a lack of optimized low precision libraries. In this paper we introduce a work flow to quickly generate high performance low precision deep learning operators for arbitrary precision that target multiple CPU architectures and include optimizations such as memory tiling and vectorization. We present an extensive case study on low power ARM Cortex-A53 CPU, and show how we can generate 1-bit, 2-bit convolutions with speedups up to 16x over an optimized 16-bit integer baseline and 2.3x better than handwritten implementations., Comment: 10 pages, 11 figures

À l’occasion de l’année Janusz Korczak, le Centre de recherche sur l’État et la Constitution (UCL), le Centre interdisciplinaire des droits de l’enfant (UCL) et l’Institut polonais - Service culturel de l’Ambassade de la République de Pologne à Bruxelles ont organisé un colloque dont le but était tout à la fois d’explorer la vision de l’enfant présente dans différents dispositifs normatifs et l’engagement de l’État à respecter l’enfant et ses droits. Ce travail a été réalisé autour de trois thèmes : l’enfant dans sa famille, l’enfant à l’école et l’enfant placé ou privé de liberté. Ils font chacun l’objet de trois rapports. Les deux premiers sont de nature juridique. Leur objet est d’examiner tant les standards internationaux que des situations nationales au regard de cette notion propre à Janusz Korczak qu’est le droit de l’enfant au respect. Le troisième rapport, rédigé par un praticien ou acteur de terrain, a pour objectif de relever les difficultés issues de la pratique qui devraient interpeller le juriste. Les contributions montrent en particulier quel regard international, orienté droits de l’homme, est porté sur des situations nationales. Elles mettent aussi en exergue des décisions ou des dispositifs innovants.

International audience; 10 The porcine ligands of human CD2 remain unknown in xenotransplantation despite being an important pathway of T cell 11 costimulation. Of the two main candidates, i.e., CD48 and CD58, the cDNA of the most likely ligand poCD58 was cloned from 12 CD48-negative endothelial cells costimulating human CD4 þ T cells through the CD2 pathway. The deduced protein sequence is 244 13 residues long and is 43% homologous to the human sequence. Based on similarity between porcine and human CD58 external V-set 14 Ig-type domains, a structural model of poCD58-huCD2 interaction was built. Most of the charged residues located at the interface 15 with huCD2 are highly conserved. Six putative hydrogen bonds between poCD58 and huCD2 were identified; five involve the same 16 residues as in the syngeneic combination while the sixth is formed between an additional tyrosine in poCD58 and Arg48 in huCD2, 17 increasing the complementarity between the two molecules. These structural data will help us to develop poCD58 blocking agents 18 for xenotransplantation. 19 21 When pig to human xenotransplantation was con-22 sidered in the early 1990s, the problem of hyperacute 23 rejection was first investigated and recently solved by 24 using organs from pig transgenic for complement regu-25 latory proteins [3]. Cellular rejection phenomena were 26 initially underestimated, mainly because previous mur-27 ine/human xenogeneic experiments suggested that phy-28 logenetic divergence prevented the recognition of ligand/ 29 receptor and receptor/counter-receptor pairs in xenoge-30 neic systems. In particular, human T cell responses to 31 xenogeneic (murine) antigen presenting cells (APCs) 32 were considered low due to inability of costimulatory 33 receptors (CD28, CD2, and LFA-1) to be engaged by 34 the costimulatory molecules on murine APC. However, 35 phylogenetic divergence between pig and human is less 36 than between murine and human. Indeed, strong xeno-37 geneic T cell responses to pig APC were observed and 38 costimulation blocking experiments clearly demon-39 strated that human CD28 [5], CD2, and LFA-1 [9] could 40 be engaged by the putative porcine counter-receptors. 41 Characterization of porcine costimulatory molecules 42 was therefore initiated with the description of poCD86 43 (also known as B7-2) [7] and then poCD80 [19]. Sub-44 sequently, molecular, functional, and structural char-45 acterization of porcine CTLA4, one of the poCD86 46 ligands, led to the unexpected finding that this molecule 47 bound weakly to human CD86 and CD80 [18] and failed 48 to inhibit human T cell responses, when used as a 49 CTLA4-Ig fusion protein. Porcine ICAM-1 was recently 50 characterized [14] and demonstrated a low degree of 51 conservation with human ICAM-1 (41% identity at the 52 protein level). However, despite being divergent, porcine 53 ICAM-1 has kept the ability to bind human LFA-1 and 54 to transmit costimulatory signals to human T cells [9]. 55 Up to now, in xenogeneic pig-to-primate models, por-56 cine cells have demonstrated their capacity to

On enseigne dans nos universités que la caractéristique principale du mineur est son incapacité générale. Il est effectivement soumis à l'autorité parentale et "bénéficie" du régime de représentation pour tous les actes un tant soit peu importants. Ce principe est cependant de moins en moins la règle absolue. Le législateur a introduit un certain nombre d'exceptions au principe d'incapacité. Diverses juridictions ont rappelé que le régime d'incapacité est fondé sur une nécessité de protéger le mineur qui n'a pas un total discernement pour poser des actes en toute connaissance de cause. La Convention des droits de l'enfant vient également rappeler que le mineur est titulaire de droits fondamentaux, tels le respect de sa vie privée, de sa correspondance et de son domicile, la liberté de pensée, de conscience et de religion, le droit d'association,... Il a paru utile de faire le point de la question à la lumière de l'évolution de la législation et de la jurisprudence.