Italy

The Mean Teacher (MT) model of Tarvainen and Valpola has shown favorable performance on several semi- supervised benchmark datasets. MT maintains a teacher model’s weights as the exponential moving average of a student model’s weights and minimizes the divergence between their probability predictions under diverse perturbations of the inputs. However, MT is known to suffer from confirmation bias, that is, reinforcing incorrect teacher model predictions. In this work, we propose a simple yet effective method called Local Clustering (LC) to mitigate the effect of confirmation bias. In MT, each data point is considered independent of other points during training; however, data points are likely to be close to each other in feature space if they share similar features. Motivated by this, we cluster data points locally by minimizing the pairwise distance between neighboring data points in feature space. Combined with a standard classification cross-entropy objective on labeled data points, the misclassified unlabeled data points are pulled towards high-density regions of their correct class with the help of their neighbors, thus improving model performance. We demonstrate on semi-supervised benchmark datasets SVHN and CIFAR-10 that adding our LC loss to MT yields significant improvements compared to MT and performance comparable to the state of the art in semi-supervised learning.

ICPR 2020

Local Clustering with Mean Teacher for Semi-supervised learning

confirmation bias

mean teacher

consistency-based methods

semi-supervised

clustering

ICPR 2020 is the premier international conference on Pattern Recognition, gathering leading researchers from around the world to present their latest achievements in Pattern Analysis and Recognition, applied Artificial Intelligence and Machine Learning. The topic of the year is " Putting Artificial intelligence to work on Patterns". 
For the first time, ICPR has employed a two-round review process similar to journal submissions, with new papers allowed to be submitted in either the first or second round and papers submitted in the first round and not accepted allowed to be revised and re-submitted for second round review. 

In this edition of the **ICPR 2020** Conference, there are 1407 Oral/Poster presentations organized in five Tracks: 
- Artificial Intelligence, Machine Learning for Pattern Analysis; 
- Biometrics, Human Analysis and Behavior Understanding; Computer Vision, Robotics and Intelligent Systems; 
- Document and Media Analysis; and 
- Image and Signal Processing. 

The **ICPR 2020** Main Conference program also includes 7 Keynotes from prominent world-renowned scientists, 8 Tutorials, 28 Demos, 8 Challenges, 2 Panels, as well as an Industry/Sponsor Exhibition. In addition, 41 Workshops on a range of important current and emerging research topics complete the ICPR 2020 program. 


poster

Interpretable Machine Learning (IML) methods are used to gain insight into the relevance of a feature of interest for the performance of a model. Commonly used IML methods differ in whether they consider features of interest in isolation, e.g., Permutation Feature Importance (PFI), or in relation to all remaining feature variables, e.g., Conditional Feature Importance (CFI). As such, the perturbation mechanisms inherent to PFI and CFI represent extreme reference points. We introduce Relative Feature Importance (RFI), a generalization of PFI and CFI that allows for a more nuanced feature importance computation beyond the PFI versus CFI dichotomy. With RFI, the importance of a feature relative to any other subset of features can be assessed, including variables that were not available at training time. We derive general interpretation rules for RFI based on a detailed theoretical analysis of the implications of relative feature relevance, and demonstrate the method's usefulness on simulated examples.

Relative Feature Importance

In feature embedding, the recovery of associated
discriminative information in the reduced subspace is critical
for downstream classifiers. In this study, a supervised feature
embedding method is proposed inspired by the well-known word
embedding technique, word2vec. Proposed embedding method
is implemented as representative learning of rank-based neighborhoods.
The notion of context words in word2vec is extended
into neighboring instances within a given window. Neighborship
is defined using ranks of instances rather than their values so
that regions with different densities are captured properly. Each
sample is represented by a unique one-hot vector whereas its
neighbors are encoded by several two-hot vectors. The two-hot
vectors are identical for neighboring samples of the same class. A
feed-forward neural network with a continuous projection layer,
then learns the mapping from one-hot vectors to multiple two-hot
vectors. The hidden layer determines the reduced subspace for
the train samples. The obtained transformation is then applied
on test data to find a lower-dimensional representation. Proposed
method is tested in classification problems on 10 UCI data
sets. Experimental results confirm that the proposed method is
effective in finding a discriminative representation of the features
and outperforms several supervised

Supervised Feature Embedding for Classification by Learning Rank-based Neighborhoods

Context is known to be one of crucial factors effecting the performance improvement of semantic segmentation. However, state-of-the-art segmentation models built upon fully convolutional networks are inherently weak in encoding contextual information because of stacked local operations such as convolution and pooling. Failing to capture context leads to inferior segmentation performance. Despite many context modules have been proposed to relieve this problem, they still operate in a local manner or use the same contextual information in different positions (due to upsampling). In this paper, we introduce the idea of Multi-Direction Convolution (MDC)—a novel operator capable of encoding rich contextual information. This operator is inspired by an observation that the standard convolution only slides along the spatial dimension (x, y direction) where the channel dimension (z direction) is fixed, which renders slow growth of the receptive field (RF). If considering the channel-fixed convolution to be one-direction, MDC is multi-direction in the sense that MDC slides along both spatial and channel dimensions, i.e., it slides along x, y when z is fixed, along x, z when y is fixed, and along y, z when x is fixed. In this way, MDC is able to encode rich contextual information with the fast increase of the RF. Compared to existing context modules, the encoded context is position-sensitive because no upsampling is required. MDC is also efficient and easy to implement. It can be implemented with few standard convolution layers with permutation. We show through extensive experiments that MDC effectively and selectively enlarges the RF and outperforms existing contextual modules on two standard benchmarks, including Cityscapes and PASCAL VOC2012.

Multi-Direction Convolution for Semantic Segmentation

We tackle depth completion task. Based on our assumption, color image and depth map share some structural information, we proposed DEN network that disentangle and exchange the structural information of color and depth domain to accomplish depth completion. In this presentation, we'll show the mechanism and design ideas of our method.

DEN: Disentangling and Exchanging Network for Depth Completion

This presentation briefly introduces our motivation and method of utilizing temporal features in dynamic point cloud segmentation. It also presents the experiment results, including comparison to other methods and ablation study on the effect of temporal features.

Human Segmentation with Dynamic LiDAR Data

An input space is a set of all the possible inputs for a neural network. An element or a group of elements in the input space can easily be understood by projecting them on their original forms. Even though Piece-wise Linear Neural Networks (PLNNs) are a nonlinear system in general, a PLNN can also be expressed in terms of linear constraints because the Rectified Linear Units (ReLU) function is a piece-wise linear function. A PLNN divides the input space into disjoint linear regions. We proved that all components of the outputs are continuous at the boundary between two different adjacent regions. This continuity implies that the boundary corresponding to a unit itself should be continuous regardless of the regions. Furthermore, we also obtained the boundaries of a single-class region, which has the same predicted classes in the interior of the region. Finally, we suggested that the point-wise robustness of a neural network can be calculated by investigating the boundaries of linear regions and the single-class regions. We obtained adversarial examples in which Euclidean distances from original inputs are less than 0.01 pixels.

Boundaries of Single-Class Regions in the Input Space of Piece-Wise Linear Neural Networks

High-order statistics have been proven useful in the framework of Convolutional Neural Networks (CNN) for a variety of computer vision tasks. 
In this paper, we propose to exploit high-order statistics in the framework of Recurrent Neural Networks (RNN) for skeleton-based hand gesture recognition. 
Our method is based on the Statistical Recurrent Units (SRU), an un-gated architecture that has been introduced as an alternative model for Long-Short Term Memory (LSTM) and Gate Recurrent Unit (GRU). 
The SRU captures sequential information by generating recurrent statistics 
that depend on a context of previously seen data and by computing moving averages at different scales. The integration of high-order statistics in the SRU significantly improves the performance of the original one, resulting in a model that is competitive to state-of-the-art methods on the Dynamic Hand Gesture (DHG) dataset,and outperforms them on the First-Person Hand Action (FPHA) dataset.

Learning Recurrent High-order Statistics
for Skeleton-based Hand Gesture Recognition

Action recognition in continuous video streams is a growing field since the past few years. Deep learning techniques and in particular Convolutional Neural Networks (CNNs) achieved good results in this topic. However, intrinsic CNNs limitations begin to cap the results since 2D CNN cannot capture temporal information and 3D CNN are too much resource demanding for real-time applications. Capsule Network, evolution of CNN, already proves its interesting benefits on small and low informational datasets like MNIST but yet its true potential has not emerged. In this paper we tackle the action recognition problem by proposing a new architecture combining Temporal Shift module over deep Capsule Network. Temporal Shift module permits us to insert temporal information over 2D Capsule Network with a zero computational cost to conserve the lightness of 2D capsules and their ability to connect spatial features. Our proposed approach outperforms or brings near state-of-the-art results on color and depth information on public datasets like First Person Hand Action and DHG 14/28 with a number of parameters 10 to 40 times less than existing approaches.

2D Deep Video Capsule Network with Temporal Shit for Action Recognition

We propose to unify two trends in unsupervised person re-ID: clustering & fine-tuning and adversarial learning. Since target data is distributed across different camera viewpoints, we propose to model each camera as an independent domain, and aim to learn domain-independent features. Straightforward adversarial learning yields negative transfer, we thus introduce a conditioning strategy to mitigate this undesirable effect. We evaluate the proposed architecture on two state-of-the-art clustering-based unsupervised person re-identification (re-ID) methods and three different data sets and set the new state-of-the-art performance on all three of them.

CANU-ReID: A Conditional Adversarial Network for Unsupervised person Re-IDentification

Visual voice activity detection (V-VAD) uses visual features to predict whether a person is speaking or not. VVAD is useful whenever audio VAD (A-VAD) is inefficient either because the acoustic signal is difficult to analyze or because it is simply missing. We propose two deep architectures for V-VAD, one based on facial landmarks and one based on optical flow. Moreover, available datasets, used for learning and for testing VVAD, lack content variability. We introduce a novel methodology to automatically create and annotate very large datasets inthe-wild – WildVVAD – based on combining A-VAD with face detection and tracking. A thorough empirical evaluation showsthe advantage of training the proposed deep V-VAD models with this dataset.

Premium content

Local Clustering with Mean Teacher for Semi-supervised learning

Speakers

Downloads

Next from ICPR 2020

Relative Feature Importance

Similar lecture

Local Clustering with Mean Teacher for Semi-supervised learning