Most successful approaches to estimate the 6D pose of an object typically train a neural network by supervising the learning with annotated poses in real world images. These annotations are generally expensive to obtain and a common workaround is to generate and train on synthetic scenes, with the drawback of limited generalisation when the model is deployed in the real world. In this work, a two-stage 6D object pose estimator framework that can be applied on top of existing neural-network-based approaches and that does not require pose annotations on real images is proposed. The first self-supervised stage enforces the pose consistency between rendered predictions and real input images, narrowing the gap between the two domains. The second stage fine-tunes the previously trained model by enforcing the photometric consistency between pairs of different object views, where one image is warped and aligned to match the view of the other and thus enabling their comparison. In the absence of both real image annotations and depth information, applying the proposed framework on top of two recent approaches results in state-of-the-art performance when compared to methods trained only on synthetic data, domain adaptation baselines and a concurrent self-supervised approach on LINEMOD, LINEMOD OCCLUSION and HomebrewedDB datasets.
Weight pruning of deep neural networks (DNNs) has been proposed to satisfy the limited storage and computing capability of mobile edge devices. However, previous pruning methods mainly focus on reducing the model size and/or improving performance without considering the privacy of user data. To mitigate this concern, we propose a privacy-preserving-oriented pruning and mobile acceleration framework that does not require the private training dataset. At the algorithm level of the proposed framework, a systematic weight pruning technique based on the alternating direction method of multipliers (ADMM) is designed to iteratively solve the pattern-based pruning problem for each layer with randomly generated synthetic data. In addition, corresponding optimizations at the compiler level are leveraged for inference accelerations on devices. With the proposed framework, users could avoid the time-consuming pruning process for non-experts and directly benefit from compressed models. Experimental results show that the proposed framework outperforms three state-of-art end-to-end DNN frameworks, i.e., TensorFlow-Lite, TVM, and MNN, with speedup up to 4.2X, 2.5X, and 2.0X, respectively, with almost no accuracy loss, while preserving data privacy.
Contextual embeddings, such as ELMo and BERT, move beyond global word representations like Word2Vec and achieve ground-breaking performance on a wide range of natural language processing tasks. Contextual embeddings assign each word a representation based on its context, thereby capturing uses of words across varied contexts and encoding knowledge that transfers across languages. In this survey, we review existing contextual embedding models, cross-lingual polyglot pre-training, the application of contextual embeddings in downstream tasks, model compression, and model analyses.
Graph convolution network (GCN) attracts intensive research interest with broad applications. While existing work mainly focused on designing novel GCN architectures for better performance, few of them studied a practical yet challenging problem: How to learn GCNs from data with extremely limited annotation? In this paper, we propose a new learning method by sampling strategy and model compression to overcome this challenge. Our approach has multifold advantages: 1) the adaptive sampling strategy largely suppresses the GCN training deviation over uniform sampling; 2) compressed GCN-based methods with a smaller scale of parameters need fewer labeled data to train; 3) the smaller scale of training data is beneficial to reduce the human resource cost to label them. We choose six popular GCN baselines and conduct extensive experiments on three real-world datasets. The results show that by applying our method, all GCN baselines cut down the annotation requirement by as much as 90$\%$ and compress the scale of parameters more than 6$\times$ without sacrificing their strong performance. It verifies that the training method could extend the existing semi-supervised GCN-based methods to the scenarios with the extremely small scale of labeled data.
This paper addresses the problem of model compression via knowledge distillation. To this end, we propose a new knowledge distillation method based on transferring feature statistics, specifically the channel-wise mean and variance, from the teacher to the student. Our method goes beyond the standard way of enforcing the mean and variance of the student to be similar to those of the teacher through an $L_2$ loss, which we found it to be of limited effectiveness. Specifically, we propose a new loss based on adaptive instance normalization to effectively transfer the feature statistics. The main idea is to transfer the learned statistics back to the teacher via adaptive instance normalization (conditioned on the student) and let the teacher network "evaluate" via a loss whether the statistics learned by the student are reliably transferred. We show that our distillation method outperforms other state-of-the-art distillation methods over a large set of experimental settings including different (a) network architectures, (b) teacher-student capacities, (c) datasets, and (d) domains.
There is a need for an on-the-fly computational process with very low performance system such as system-on-chip (SoC) and embedded device etc. This paper presents pacemaker knowledge distillation as intermediate ensemble teacher to use convolutional neural network in these systems. For on-the-fly system, we consider student model using 1xN shape on-the-fly filter and teacher model using normal NxN shape filter. We note three points about training student model, caused by applying on-the-fly filter. First, same depth but unavoidable thin model compression. Second, the large capacity gap and parameter size gap due to only the horizontal field must be selected not the vertical receptive. Third, the performance instability and degradation of direct distilling. To solve these problems, we propose intermediate teacher, named pacemaker, for an on-the-fly student. So, student can be trained from pacemaker and original teacher step by step. Experiments prove our proposed method make significant performance (accuracy) improvements: on CIFAR100, 5.39% increased in WRN-40-4 than conventional knowledge distillation which shows even low performance than baseline. And we solve train instability, occurred when conventional knowledge distillation was applied without proposed method, by reducing deviation range by applying proposed method pacemaker knowledge distillation.
Compressing deep neural network (DNN) models becomes a very important and necessary technique for real-world applications, such as deploying those models on mobile devices. Knowledge distillation is one of the most popular methods for model compression, and many studies have been made on developing this technique. However, those studies mainly focused on classification problems, and very few attempts have been made on regression problems, although there are many application of DNNs on regression problems. In this paper, we propose a new formalism of knowledge distillation for regression problems. First, we propose a new loss function, teacher outlier rejection loss, which rejects outliers in training samples using teacher model predictions. Second, we consider a multi-task network with two outputs: one estimates training labels which is in general contaminated by noisy labels; And the other estimates teacher model's output which is expected to modify the noise labels following the memorization effects. By considering the multi-task network, training of the feature extraction of student models becomes more effective, and it allows us to obtain a better student model than one trained from scratch. We performed comprehensive evaluation with one simple toy model: sinusoidal function, and two open datasets: MPIIGaze, and Multi-PIE. Our results show consistent improvement in accuracy regardless of the annotation error level in the datasets.
Pre-trained Transformer-based models have achieved state-of-the-art performance for various Natural Language Processing (NLP) tasks. However, these models often have billions of parameters, and, thus, are too resource-hungry and computation-intensive to suit low-capability devices or applications with strict latency requirements. One potential remedy for this is model compression, which has attracted a lot of research attention. Here, we summarize the research in compressing Transformers, focusing on the especially popular BERT model. In particular, we survey the state of the art in compression for BERT, we clarify the current best practices for compressing large-scale Transformer models, and we provide insights into the workings of various methods. Our categorization and analysis also shed light on promising future research directions for achieving lightweight, accurate, and generic NLP models.
The enormous inference cost of deep neural networks can be scaled down by network compression. Pruning is one of the predominant approaches used for deep network compression. However, existing pruning techniques have one or more of the following limitations: 1) Additional energy cost on top of the compute heavy training stage due to pruning and fine-tuning stages, 2) Layer-wise pruning based on the statistics of a particular, ignoring the effect of error propagation in the network, 3) Lack of an efficient estimate for determining the important channels globally, 4) Unstructured pruning requires specialized hardware for effective use. To address all the above issues, we present a simple-yet-effective gradual channel pruning while training methodology using a novel data-driven metric referred to as feature relevance score. The proposed technique gets rid of the additional retraining cycles by pruning the least important channels in a structured fashion at fixed intervals during the actual training phase. Feature relevance scores help in efficiently evaluating the contribution of each channel towards the discriminative power of the network. We demonstrate the effectiveness of the proposed methodology on architectures such as VGG and ResNet using datasets such as CIFAR-10, CIFAR-100 and ImageNet, and successfully achieve significant model compression while trading off less than $1\%$ accuracy. Notably on CIFAR-10 dataset trained on ResNet-110, our approach achieves $2.4\times$ compression and a $56\%$ reduction in FLOPs with an accuracy drop of $0.01\%$ compared to the unpruned network.
Knowledge distillation (KD) is one of the most potent ways for model compression. The key idea is to transfer the knowledge from a deep teacher model (T) to a shallower student (S). However, existing methods suffer from performance degradation due to the substantial gap between the learning capacities of S and T. To remedy this problem, this work proposes Residual Knowledge Distillation (RKD), which further distills the knowledge by introducing an assistant (A). Specifically, S is trained to mimic the feature maps of T, and A aids this process by learning the residual error between them. In this way, S and A complement with each other to get better knowledge from T. Furthermore, we devise an effective method to derive S and A from a given model without increasing the total computational cost. Extensive experiments show that our approach achieves appealing results on popular classification datasets, CIFAR-100 and ImageNet, surpassing state-of-the-art methods.
Convolutional Neural Networks (CNN) are becoming a common presence in many applications and services, due to their superior recognition accuracy. They are increasingly being used on mobile devices, many times just by porting large models designed for server space, although several model compression techniques have been considered. One model compression technique intended to reduce computations is channel pruning. Mobile and embedded systems now have GPUs which are ideal for the parallel computations of neural networks and for their lower energy cost per operation. Specialized libraries perform these neural network computations through highly optimized routines. As we find in our experiments, these libraries are optimized for the most common network shapes, making uninstructed channel pruning inefficient. We evaluate higher level libraries, which analyze the input characteristics of a convolutional layer, based on which they produce optimized OpenCL (Arm Compute Library and TVM) and CUDA (cuDNN) code. However, in reality, these characteristics and subsequent choices intended for optimization can have the opposite effect. We show that a reduction in the number of convolutional channels, pruning 12% of the initial size, is in some cases detrimental to performance, leading to 2x slowdown. On the other hand, we also find examples where performance-aware pruning achieves the intended results, with performance speedups of 3x with cuDNN and above 10x with Arm Compute Library and TVM. Our findings expose the need for hardware-instructed neural network pruning.
We propose and evaluate a novel procedure for training multiple Transformers with tied parameters which compresses multiple models into one enabling the dynamic choice of the number of encoder and decoder layers during decoding. In sequence-to-sequence modeling, typically, the output of the last layer of the N-layer encoder is fed to the M-layer decoder, and the output of the last decoder layer is used to compute loss. Instead, our method computes a single loss consisting of NxM losses, where each loss is computed from the output of one of the M decoder layers connected to one of the N encoder layers. Such a model subsumes NxM models with different number of encoder and decoder layers, and can be used for decoding with fewer than the maximum number of encoder and decoder layers. We then propose a mechanism to choose a priori the number of encoder and decoder layers for faster decoding, and also explore recurrent stacking of layers and knowledge distillation for model compression. We present a cost-benefit analysis of applying the proposed approaches for neural machine translation and show that they reduce decoding costs while preserving translation quality.
We propose a novel algorithm for quantizing continuous latent representations in trained models. Our approach applies to deep probabilistic models, such as variational autoencoders (VAEs), and enables both data and model compression. Unlike current end-to-end neural compression methods that cater the model to a fixed quantization scheme, our algorithm separates model design and training from quantization. Consequently, our algorithm enables "plug-and-play" compression with variable rate-distortion trade-off, using a single trained model. Our algorithm can be seen as a novel extension of arithmetic coding to the continuous domain, and uses adaptive quantization accuracy based on estimates of posterior uncertainty. Our experimental results demonstrate the importance of taking into account posterior uncertainties, and show that image compression with the proposed algorithm outperforms JPEG over a wide range of bit rates using only a single standard VAE. Further experiments on Bayesian neural word embeddings demonstrate the versatility of the proposed method.
Weight pruning is a powerful technique to realize model compression. We propose PCNN, a fine-grained regular 1D pruning method. A novel index format called Sparsity Pattern Mask (SPM) is presented to encode the sparsity in PCNN. Leveraging SPM with limited pruning patterns and non-zero sequences with equal length, PCNN can be efficiently employed in hardware. Evaluated on VGG-16 and ResNet-18, our PCNN achieves the compression rate up to 8.4X with only 0.2% accuracy loss. We also implement a pattern-aware architecture in 55nm process, achieving up to 9.0X speedup and 28.39 TOPS/W efficiency with only 3.1% on-chip memory overhead of indices.
Knowledge Distillation (KD) is a model-agnostic technique to improve model quality while having a fixed capacity budget. It is a commonly used technique for model compression, where a larger capacity teacher model with better quality is used to train a more compact student model with better inference efficiency. Through distillation, one hopes to benefit from student's compactness, without sacrificing too much on model quality. Despite the large success of knowledge distillation, better understanding of how it benefits student model's training dynamics remains under-explored. In this paper, we categorize teacher's knowledge into three hierarchical levels and study its effects on knowledge distillation: (1) knowledge of the `universe', where KD brings a regularization effect through label smoothing; (2) domain knowledge, where teacher injects class relationships prior to student's logit layer geometry; and (3) instance specific knowledge, where teacher rescales student model's per-instance gradients based on its measurement on the event difficulty. Using systematic analyses and extensive empirical studies on both synthetic and real-world datasets, we confirm that the aforementioned three factors play a major role in knowledge distillation. Furthermore, based on our findings, we diagnose some of the failure cases of applying KD from recent studies.
In this paper, we propose a novel model compression approach to effectively compress BERT by progressive module replacing. Our approach first divides the original BERT into several modules and builds their compact substitutes. Then, we randomly replace the original modules with their substitutes to train the compact modules to mimic the behavior of the original modules. We progressively increase the probability of replacement through the training. In this way, our approach brings a deeper level of interaction between the original and compact models. Compared to the previous knowledge distillation approaches for BERT compression, our approach does not introduce any additional loss function. Our approach outperforms existing knowledge distillation approaches on GLUE benchmark, showing a new perspective of model compression.
The increasing computational and memory complexities of deep neural networks have made it difficult to deploy them on low-resource electronic devices (e.g., mobile phones, tablets, wearables). Practitioners have developed numerous model compression methods to address these concerns, but few have condensed input representations themselves. In this work, we propose a fast, accurate, and lightweight convolutional representation that can be swapped into any neural model and compressed significantly (up to 32x) with a negligible reduction in performance. In addition, we show gains over recurrent representations when considering resource-centric metrics (e.g., model file size, latency, memory usage) on a Samsung Galaxy S9.
Knowledge Distillation, as a model compression technique, has received great attention. The knowledge of a well-performed teacher is distilled to a student with a small architecture. The architecture of the small student is often chosen to be similar to their teacher's, with fewer layers or fewer channels, or both. However, even with the same number of FLOPs or parameters, the students with different architecture can achieve different generalization ability. The configuration of a student architecture requires intensive network architecture engineering. In this work, instead of designing a good student architecture manually, we propose to search for the optimal student automatically. Based on L1-norm optimization, a subgraph from the teacher network topology graph is selected as a student, the goal of which is to minimize the KL-divergence between student's and teacher's outputs. We verify the proposal on CIFAR10 and CIFAR100 datasets. The empirical experiments show that the learned student architecture achieves better performance than ones specified manually. We also visualize and understand the architecture of the found student.
Conversational agents such as Cortana, Alexa and Siri are continuously working on increasing their capabilities by adding new domains. The support of a new domain includes the design and development of a number of NLU components for domain classification, intents classification and slots tagging (including named entity recognition). Each component only performs well when trained on a large amount of labeled data. Second, these components are deployed on limited-memory devices which requires some model compression. Third, for some domains such as the health domain, it is hard to find a single training data set that covers all the required slot types. To overcome these mentioned problems, we present a multi-task transformer-based neural architecture for slot tagging. We consider the training of a slot tagger using multiple data sets covering different slot types as a multi-task learning problem. The experimental results on the biomedical domain have shown that the proposed approach outperforms the previous state-of-the-art systems for slot tagging on the different benchmark biomedical datasets in terms of (time and memory) efficiency and effectiveness. The output slot tagger can be used by the conversational agent to better identify entities in the input utterances.
We propose a Small, Accurate, and Fast Re-ID (SAFR) design for flexible vehicle re-id under a variety of compute environments such as cloud, mobile, edge, or embedded devices by only changing the re-id model backbone. Through best-fit design choices, feature extraction, training tricks, global attention, and local attention, we create a reid model design that optimizes multi-dimensionally along model size, speed, & accuracy for deployment under various memory and compute constraints. We present several variations of our flexible SAFR model: SAFR-Large for cloud-type environments with large compute resources, SAFR-Small for mobile devices with some compute constraints, and SAFR-Micro for edge devices with severe memory & compute constraints. SAFR-Large delivers state-of-the-art results with mAP 81.34 on the VeRi-776 vehicle re-id dataset (15% better than related work). SAFR-Small trades a 5.2% drop in performance (mAP 77.14 on VeRi-776) for over 60% model compression and 150% speedup. SAFR-Micro, at only 6MB and 130MFLOPS, trades 6.8% drop in accuracy (mAP 75.80 on VeRi-776) for 95% compression and 33x speedup compared to SAFR-Large.