Parseval networks control spectral norms of layers, improving adversarial robustness while maintaining accuracy, by constraining weights to approximate Parseval tight frames.

Moustapha Cisse, Piotr Bojanowski, Edouard Grave et al.

Proposed an unsupervised method for learning depth and ego-motion from videos, achieving performance comparable to supervised methods on the KITTI dataset.

Tinghui Zhou, Matthew Brown, Noah Snavely et al.

Proposed Residual Attention Network achieves 3.90% error on CIFAR-10, improving accuracy by 0.6%.

Fei Wang, Mengqing Jiang, Chen Qian et al.

Proposes differentiable ray consistency (DRC) for single-view 3D reconstruction, leveraging multi-view supervision to improve accuracy.

Shubham Tulsiani, Tinghui Zhou, Alexei A. Efros et al.

Network Dissection measures CNN unit semantics with Broden and IoU; random rotation reduces unique detectors by 80%.

David Bau, Bolei Zhou, Aditya Khosla et al.

SearchQA dataset combines Jeopardy! questions with Google search snippets, revealing a significant gap between human and machine performance, advancing QA benchmarks.

Matthew Dunn, Levent Sagun, Mike Higgins et al.

MobileNets use depthwise separable convolutions with width and resolution multipliers, achieving 70.6% top-1 accuracy on ImageNet with only 4.2M parameters, 569M Mult-Adds.

Andrew G. Howard, Menglong Zhu, Bo Chen et al.

This paper evaluates Google's TPU, a custom ASIC with 65,536 8-bit MAC units achieving 92 TOPS, demonstrating superior inference performance and efficiency over CPUs and GPUs.

Norman P. Jouppi, Cliff Young, Nishant Patil et al.

AnchorNet: Weakly supervised network learning geometry-sensitive features, boosting semantic matching accuracy by 6%.

David Novotny, Diane Larlus, Andrea Vedaldi

ShapeWorld framework evaluates multimodal language understanding with detailed control and generation capabilities.

Alexander Kuhnle, Ann Copestake

Soft-NMS replaces hard thresholding with score decay based on overlap, boosting mAP by 1.7% without retraining.

Navaneeth Bodla, Bharat Singh, Rama Chellappa et al.

Pointer-generator with coverage improves factual accuracy and reduces repetition, achieving at least 2 ROUGE points gain on CNN/Daily Mail.

Abigail See, Peter J. Liu, Christopher D. Manning

Virtual Adversarial Training (VAT) uses virtual adversarial loss for regularization in supervised and semi-supervised learning, achieving top performance on SVHN and CIFAR-10.

Takeru Miyato, Shin-ichi Maeda, Masanori Koyama et al.

IRCNN embeds CNN denoisers into HQS, reaching 29.15 dB on BSD68 at σ=25 and extending to deblurring and super-resolution.

Kai Zhang, Wangmeng Zuo, Shuhang Gu et al.

Proposes a pixelwise instance segmentation method with a dynamically instantiated network, achieving state-of-the-art results at high IoU thresholds.

Anurag Arnab, Philip H. S Torr

Extends stacking to Bayesian predictive distributions, using Pareto smoothed importance sampling for efficient leave-one-out estimation, outperforming BMA and pseudo-BMA.

Yuling Yao, Aki Vehtari, Daniel Simpson et al.

Proposed a deep metric learning method combining triplet model and global structure, achieving breakthroughs on CUB-200-2011 and Cars196 datasets.

Ben Harwood, Vijay Kumar B G, Gustavo Carneiro et al.

ECM integrates emotion category embedding, internal emotion memory, and external emotion vocabulary to generate emotionally consistent responses.

Hao Zhou, Minlie Huang, Tianyang Zhang et al.

Introduces Hierarchical Surface Prediction (HSP) framework for high-resolution 256x256x256 voxel grid prediction.

Christian Häne, Shubham Tulsiani, Jitendra Malik

Risk-averse model predictive control integrates stochastic and worst-case MPC for nonlinear Markovian switching systems.

Pantelis Sopasakis, Domagoj Herceg, Alberto Bemporad et al.