检索结果-内蒙古大学图书馆

CLUTRR: A Diagnostic Benchmark for Inductive Reasoning from Text 9

CLUTRR: A Diagnostic Benchmark for Inductive Reasoning from ...

Conference on Empirical Methods in Natural Language Processing / 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)

作者： Sinha, Koustuv Sodhani, Shagun Dong, Jin Pineau, Joelle Hamilton, William L. McGill Univ Sch Comp Sci Montreal PQ Canada Univ Montreal Montreal PQ Canada Montreal Inst Learning Algorithms Mila Montreal PQ Canada Facebook AI Res FAIR Montreal PQ Canada

ISBN: (纸本)9781950737901

The recent success of natural language understanding (NLU) systems has been troubled by results highlighting the failure of these models to generalize in a systematic and robust way. In this work, we introduce a diagnostic benchmark suite, named CLUTRR, to clarify some key issues related to the robustness and systematicity of NLU systems. Motivated by classic work on inductive logic programming, CLUTRR requires that an NLU system infer kinship relations between characters in short stories. Successful performance on this task requires both extracting relationships between entities, as well as inferring the logical rules governing these relationships. CLUTRR allows us to precisely measure a model's ability for systematic generalization by evaluating on held-out combinations of logical rules, and it allows us to evaluate a model's robustness by adding curated noise facts. Our empirical results highlight a substantial performance gap between state-of-the-art NLU models (e.g., BERT and MAC) and a graph neural network model that works directly with symbolic inputs-with the graph-based model exhibiting both stronger generalization and greater robustness.

关键词： Graphic methods

来源：评论

学校读者我要写书评

暂无评论

On the Compound Broadcast Channel: Multiple Description Coding and Interference Decoding

引用

IEEE TRANSACTIONS ON INFORMATION THEORY 2020年第1期66卷 38-64页

作者： Benammar, Meryem Piantanida, Pablo Shamai, Shlomo Cent Supelec Dept Telecommun F-91190 Gif Sur Yvette France Inst Super Aeronaut & Espace ISAE Supaero Dept Elect Optron & Signal Proc F-31055 Toulouse France Univ Paris Sud French Natl Ctr Sci Res CNRS Cent Supelec F-91192 Gif Sur Yvette France Univ Montreal Montreal Inst Learning Algorithms Mila Montreal PQ H3T 1N8 Canada Technion Israel Inst Technol Dept Elect Engn IL-3200003 Haifa Israel

This work investigates the general two-user compound Broadcast Channel (BC) in which an encoder wishes to transmit two private messages W-1 and W-2 to two receivers while being oblivious to the actual channel realizations controlling the communication. The focus is on the characterization of the largest achievable rate region by resorting to more involved encoding and decoding techniques than the usual coding schemes of the standard BC. Involved decoding schemes are first explored, and an achievable rate region is derived based on the principle of Interference Decoding (ID), in which each receiver decodes its intended message and chooses to (non-uniquely) decode, or not, the interfering non-itended message. This decoding scheme is shown to be capacity achieving for a class of non-trivial compound BEC/BSC broadcast channels while the worst-case of Marton's inner bound-based on No Interference Decoding (NID)-fails to achieve the capacity region. Involved encoding schemes are later investigated, and an achievable rate region is derived based on Multiple Description (MD) coding wherin the encoder transmits a common description as well as multiple dedicated private descriptions to the many possible channel realizations of the users. It turns out that MD coding yields larger inner bounds than the single description scheme-Common Description (CD) coding-for a class of compound Multiple Input Single Output Broadcast Channels (MISO BC).

关键词： Encoding Interference Compounds Decoding MISO communication Uncertainty Receivers Channel capacity channel state information broadcast channels (BCs) compound channels multiple description coding multiple input single output broadcast channel (MISO BC)

来源：评论

学校读者我要写书评

暂无评论

Interpolation Consistency Training for Semi-Supervised learning 28

Interpolation Consistency Training for Semi-Supervised Learn...

引用

28th International Joint Conference on Artificial Intelligence

作者： Verma, Vikas Lamb, Alex Kannala, Juho Bengio, Yoshua Lopez-Paz, David Aalto Univ Espoo Finland Montreal Inst Learning Algorithms MILA Montreal PQ Canada Facebook Artificial Intelligence Res FAIR Menlo Pk CA USA

ISBN: (纸本)9780999241141

We introduce Interpolation Consistency Training (ICT), a simple and computation efficient algorithm for training Deep Neural Networks in the semi-supervised learning paradigm. ICT encourages the prediction at an interpolation of unlabeled points to be consistent with the interpolation of the predictions at those points. In classification problems, ICT moves the decision boundary to low-density regions of the data distribution. Our experiments show that ICT achieves state-of-the-art performance when applied to standard neural network architectures on the CIFAR-10 and SVHN benchmark datasets.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Fundamental Limits of Decentralized Data Shuffling

引用

IEEE TRANSACTIONS ON INFORMATION THEORY 2020年第6期66卷 3616-3637页

作者： Wan, Kai Tuninetti, Daniela Ji, Mingyue Caire, Giuseppe Piantanida, Pablo Tech Univ Berlin Elect Engn & Comp Sci Dept D-10587 Berlin Germany Univ Illinois Elect & Comp Engn Dept Chicago IL 60607 USA Univ Utah Elect & Comp Engn Dept Salt Lake City UT 84112 USA Univ Paris Sud Cent Supelec French Natl Ctr Sci Res CNRS F-91192 Gif Sur Yvette France Univ Montreal Montreal Inst Learning Algorithms MILA Montreal PQ H3T 1N8 Canada

Data shuffling of training data among different computing nodes (workers) has been identified as a core element to improve the statistical performance of modern large-scale machine learning algorithms. Data shuffling is often considered as one of the most significant bottlenecks in such systems due to the heavy communication load. Under a master-worker architecture (where a master has access to the entire dataset and only communication between the master and the workers is allowed) coding has been recently proved to considerably reduce the communication load. This work considers a different communication paradigm referred to as decentralized data shuffling, where workers are allowed to communicate with one another via a shared link. The decentralized data shuffling problem has two phases: workers communicate with each other during the data shuffling phase, and then workers update their stored content during the storage phase. The main challenge is to derive novel converse bounds and achievable schemes for decentralized data shuffling by considering the asymmetry of the workers' storages (i.e., workers are constrained to store different files in their storages based on the problem setting), in order to characterize the fundamental limits of this problem. For the case of uncoded storage (i.e., each worker directly stores a subset of bits of the dataset), this paper proposes converse and achievable bounds (based on distributed interference alignment and distributed clique-covering strategies) that are within a factor of 3/2 of one another. The proposed schemes are also exactly optimal under the constraint of uncoded storage for either large storage size or at most four workers in the system.

关键词： Distributed databases Machine learning algorithms Device-to-device communication Computational modeling Topology Interference Protocols Decentralized data shuffling uncoded storage distributed clique covering

来源：评论

学校读者我要写书评

暂无评论

Manifold Mixup: Better Representations by Interpolating Hidden States 36

Manifold Mixup: Better Representations by Interpolating Hidd...

引用

36th International Conference on Machine learning (ICML)

作者： Verma, Vikas Lamb, Alex Beckham, Christopher Najafi, Amir Mitliagkas, Ioannis Lopez-Paz, David Bengio, Yoshua Aalto Univ Espoo Finland Montreal Inst Learning Algorithms MILA Montreal PQ Canada Sharif Univ Technol Tehran Iran Facebook Res Menlo Pk CA USA

ISBN: (纸本)9781510886988

Deep neural networks excel at learning the training data, but often provide incorrect and confident predictions when evaluated on slightly different test examples. This includes distribution shifts, outliers, and adversarial examples. To address these issues, we propose Manifold Mixup, a simple regularizer that encourages neural networks to predict less confidently on interpolations of hidden representations. Manifold Mixup leverages semantic interpolations as additional training signal, obtaining neural networks with smoother decision boundaries at multiple levels of representation. As a result, neural networks trained with Manifold Mixup learn flatter class-representations, that is, with fewer directions of variance. We prove theory on why this flattening happens under ideal conditions, validate it empirically on practical situations, and connect it to the previous works on information theory and generalization. In spite of incurring no significant computation and being implemented in a few lines of code, Manifold Mixup improves strong baselines in supervised learning, robustness to single-step adversarial attacks, and test log-likelihood.

关键词： Interpolation

来源：评论

学校读者我要写书评

暂无评论

Two types of human TCR differentially regulate reactivity to self and non-self antigens

引用

ISCIENCE 2022年第9期25卷 104968页

作者： Trofimov, Assya Brouillard, Philippe Larouche, Jean-David Seguin, Jonathan Laverdure, Jean-Philippe Brasey, Ann Ehx, Gregory Roy, Denis-Claude Busque, Lambert Lachance, Silvy Lemieux, Sebastien Perreault, Claude Univ Montreal Inst Res Immunol & Canc IRIC Montreal PQ H3C 3J7 Canada Univ Montreal Dept Comp Sci & Res Operat Montreal PQ H3C 3J7 Canada Quebec Inst Learning Algorithms Mila Montreal PQ H2S 3H1 Canada Univ Montreal Dept Med Montreal PQ H3C 3J7 Canada Maisonneuve Rosemt Hosp Montreal PQ H1T 2M4 Canada Univ Liege Currently Interdisciplinary Cluster Appl Geno Pro B-4000 Liege Belgium Fred Hutchinson Canc Ctr Seattle WA 98109 USA Univ Washington Dept Phys Seattle WA 98195 USA Univ Montreal Dept Biochem Montreal PQ H3C 3J7 Canada

Based on analyses of TCR sequences from over 1,000 individuals, we report that the TCR repertoire is composed of two ontogenically and functionally distinct types of TCRs. Their production is regulated by variations in thymic output and terminal deoxynucleotidyl transferase (TDT) activity. Neonatal TCRs derived from TDT-negative progenitors persist throughout life, are highly shared among subjects, and are reported as disease-associated. Thus, 10%-30% of most frequent cord blood TCRs are associated with common pathogens and autoantigens. TDT-dependent TCRs present distinct structural features and are less shared among subjects. TDT-dependent TCRs are produced in maximal numbers during infancy when thymic output and TDT activity reach a summit, are more abundant in subjects with AIRE mutations, and seem to play a dominant role in graft-versus-host disease. Factors decreasing thymic output (age, male sex) negatively impact TCR diversity. Males compensate for their lower repertoire diversity via hyperexpansion of selected TCR clonotypes.

关键词： Immune response Immunity Immunology

来源：评论

学校读者我要写书评

暂无评论

New Results on Testing Against Independence with Rate-Limited Constraints 7

New Results on Testing Against Independence with Rate-Limite...

引用

7th IEEE Global Conference on Signal and Information Processing (IEEE GlobalSIP)

作者： Espinosa, Sebastian Silva, Jorge F. Piantanida, Pablo Univ Chile Dept Elect Engn Santiago Chile Univ Paris Sud CNRS Cent Supelec Lab Signaux & Syst L2S Orsay France Univ Montreal Montreal Inst Learning Algorithms MILA Montreal PQ Canada

ISBN: (纸本)9781728127231

This work studies error exponent limits in hypothesis testing (HT) in a distributed scenario with partial communication constraints. We derive general conditions on the Type I error restriction under which the error exponent of the optimal Type II error has a closed-form characterization for the task of testing against independence. We show that the error exponent is preserved for a family of decreasing Type I error restrictions. Complementing this analysis, new expressions are derived to bound the optimal Type II error probability for a finite number of observations. These bounds shed light about the velocity at which error exponent limits are attained with the number of samples.

关键词： Distributed hypothesis testing multi-terminal source coding error exponent information bottleneck

来源：评论

学校读者我要写书评

暂无评论

vGraph: A Generative Model for Joint Community Detection and Node Representation learning

vGraph: A Generative Model for Joint Community Detection and...

引用

33rd Conference on Neural Information Processing Systems (NeurIPS)

作者： Sun, Fan-Yun Qu, Meng Hoffmann, Jordan Huang, Chin-Wei Tang, Jian Natl Taiwan Univ Taipei Taiwan Mila Quebec Inst Learning Algorithms Montreal PQ Canada Harvard Univ Cambridge MA 02138 USA Element AI Montreal PQ Canada HEC Montreal Montreal PQ Canada CIFAR AI Res Chair Montreal PQ Canada

This paper focuses on two fundamental tasks of graph analysis: community detection and node representation learning, which capture the global and local structures of graphs, respectively. In the current literature, these two tasks are usually independently studied while they are actually highly correlated. We propose a probabilistic generative model called vGraph to learn community membership and node representation collaboratively. Specifically, we assume that each node can be represented as a mixture of communities, and each community is defined as a multinomial distribution over nodes. Both the mixing coefficients and the community distribution are parameterized by the low-dimensional representations of the nodes and communities. We designed an effective variational inference algorithm which regularizes the community membership of neighboring nodes to be similar in the latent space. Experimental results on multiple real-world graphs show that vGraph is very effective in both community detection and node representation learning, outperforming many competitive baselines in both tasks. We show that the framework of vGraph is quite flexible and can be easily extended to detect hierarchical communities.

关键词： Inference engines

来源：评论

学校读者我要写书评

暂无评论

An End-to-End Neighborhood-based Interaction Model for Knowledge-enhanced Recommendation 1

An End-to-End Neighborhood-based Interaction Model for Knowl...

引用

1st International Workshop on Deep learning Practice for High-Dimensional Sparse Data with KDD (DLP KDD)

作者： Qu, Yanru Bai, Ting Zhang, Weinan Nie, Jianyun Tang, Jian Shanghai Jiao Tong Univ Shanghai Peoples R China Beijing Univ Posts & Telecommun Beijing Peoples R China Renmin Univ China Beijing Peoples R China Univ Montreal Montreal PQ Canada Mila Quebec Inst Learning Algorithms Montreal PQ Canada HEC Montreal Montreal PQ Canada CIFAR AI Res Chair Edmonton AB Canada

ISBN: (纸本)9781450367837

This paper studies graph-based recommendation, where an interaction graph is built from historical responses and is leveraged to alleviate data sparsity and cold start problems. We reveal an early summarization problem in previous graph-based models, and propose Neighborhood Interaction (NI) model to capture each neighbor pair (between user-side and item-side) distinctively. NI model is more expressive and captures more complicated structural patterns behind user-item interactions. To enrich the neighborhood information, we also introduce Graph Neural Networks (GNNs) and Knowledge Graphs (KGs) to NI, resulting an end-to-end model, namely Knowledge-enhanced Neighborhood Interaction (KNI). Our experiments on 4 real world datasets show that, compared with state-of-the-art feature-based, meta path-based, and KG-based recommendation models, KNI achieves superior performance in click-through rate prediction (1.1%-8.4% absolute AUC improvements) and outperforms by a wide margin in top-N recommendation.

关键词： Knowledge Graph Neighborhood-based Interaction Collaborative Recommendation

来源：评论

学校读者我要写书评

暂无评论

Augmented CycleGAN: learning Many-to-Many Mappings from Unpaired Data 35

Augmented CycleGAN: Learning Many-to-Many Mappings from Unpa...

引用

35th International Conference on Machine learning (ICML)

作者： Almahairi, Amjad Rajeswar, Sai Sordoni, Alessandro Bachman, Philip Courville, Aaron Montreal Inst Learning Algorithms MILA Montreal PQ Canada Microsoft Res Montreal Montreal PQ Canada MSR Montreal Montreal PQ Canada

ISBN: (纸本)9781510867963

learning inter-domain mappings from unpaired data can improve performance in structured prediction tasks, such as image segmentation, by reducing the need for paired data. CycleGAN was recently proposed for this problem, but critically assumes the underlying inter-domain mapping is approximately deterministic and one-to-one. This assumption renders the model ineffective for tasks requiring flexible, many-to-many mappings. We propose a new model, called Augmented CycleGAN, which learns many-to-many mappings between domains. We examine Augmented CycleGAN qualitatively and quantitatively on several image datasets.

关键词： Image segmentation

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：