Lifelong or continual learning remains to be a challenge for artificial neural network, as it is required to be both stable for preservation of old knowledge and plastic for acquisition of new knowledge. It is common to see previous experience get overwritten, which leads to the well-known issue of catastrophic forgetting, especially in the scenario of class-incremental learning (Class-IL). Recently, many lifelong learning methods have been proposed to avoid catastrophic forgetting. However, models which learn without replay of the input data, would encounter another problem which has been ignored, and we refer to it as prior information loss (PIL). In training procedure of Class-IL, as the model has no knowledge about following tasks, it would only extract features necessary for tasks learned so far, whose information is insufficient for joint classification. In this paper, our empirical results on several image datasets show that PIL limits the performance of current state-of-the-art method for Class-IL, the orthogonal weights modification (OWM) algorithm. Furthermore, we propose to combine self-supervised learning, which can provide effective representations without requiring labels, with Class-IL to partly get around this problem. Experiments show superiority of proposed method to OWM, as well as other strong baselines.

Abstract: The algorithm of brain learning and memory is still undetermined. The backpropagation algorithm of artificial neural networks was thought not suitable for brain cortex, and there is a lack of algorithm for memory engram. We designed a brain version of backpropagation algorithm, which are biologically plausible and could be implemented with virtual neurons to complete image classification task. An encoding algorithm that can automatically allocate engram cells is proposed, which is an algorithm implementation for memory engram theory, and could simulate how hippocampus achieve fast associative memory. The role of the LTP and LTD in the cerebellum is also explained in algorithm level. Our results proposed a method for the brain to deploy backpropagation algorithm, and sparse coding method for memory engram theory. Recently, artificial intelligence based on deep neural networks has made great progress in games, medical and other fields (1,2). Backpropagation (BP) as the optimization algorithm of artificial neural network (3), is considered biologically implausible (4). The learning algorithm of biological neural network is still uncertain. The framework Neural Gradient Representation by Activity Differences (NGRAD) was proposed to be a solution for brain to implement BP algorithm, but there is currently a lack of ideal experimental results (5). It is still a mystery that whether the learning algorithm of the cerebral cortex is backpropagation or other algorithms.

Diabetic retinopathy (DR) is a retinal microvascular condition that emerges in diabetic patients. DR will continue to be a leading cause of blindness worldwide, with a predicted 191.0 million globally diagnosed patients in 2030. Microaneurysms, hemorrhages, exudates, and cotton wool spots are common signs of DR. However, they can be small and hard for human eyes to detect. Early detection of DR is crucial for effective clinical treatment. Existing methods to classify images require much time for feature extraction and selection, and are limited in their performance. Convolutional Neural Networks (CNNs), as an emerging deep learning (DL) method, have proven their potential in image classification tasks. In this paper, comprehensive experimental studies of implementing state-of-the-art CNNs for the detection and classification of DR are conducted in order to determine the top performing classifiers for the task. Five CNN classifiers, namely Inception-V3, VGG19, VGG16, ResNet50, and InceptionResNetV2, are evaluated through experiments. They categorize medical images into five different classes based on DR severity. Data augmentation and transfer learning techniques are applied since annotated medical images are limited and imbalanced. Experimental results indicate that the ResNet50 classifier has top performance for binary classification and that the InceptionResNetV2 classifier has top performance for multi-class DR classification.

Deep neural networks (DNNs) perform very well when trained on a large amount of data [1], but large SSVEP datasets are not commonly available for open use. Our way to overcome this problem was to employ data augmentation and transfer learning techniques to train the DNNs, as both are known to improve the performances of DNNs on smaller datasets [2]. We started with an open SSVEP dataset [3], which we consider to be large in comparison with other open databases. The electroencephalography (EEG) signals where transformed into images, specifically spectrograms, using the shorttime Fourier transform (STFT). By doing so, we take advantage of the ability of convolutional DNN in classifying images, which is well documented [1]. The neural network used in this study [4] is a DCNN based on the original VGG [5].

Current brain white matter fiber tracking techniques show a number of problems, including: generating large proportions of streamlines that do not accurately describe the underlying anatomy; extracting streamlines that are not supported by the underlying diffusion signal; and under-representing some fiber populations, among others. In this paper, we describe a novel unsupervised learning method to filter streamlines from diffusion MRI tractography, and hence, to obtain more reliable tractograms. We show that a convolutional neural network autoencoder provides a straightforward and elegant way to learn a robust representation of brain streamlines, which can be used to filter undesired samples with a nearest neighbor algorithm. Our method, dubbed FINTA (Filtering in Tractography using Autoencoders) comes with several key advantages: training does not need labeled data, as it uses raw tractograms, it is fast and easily reproducible, it does not rely on the input diffusion MRI data, and thus, does not suffer from domain adaptation issues. We demonstrate the ability of FINTA to discriminate between "plausible" and "implausible" streamlines as well as to recover individual streamline group instances from a raw tractogram, from both synthetic and real human brain diffusion MRI tractography data, including partial tractograms. Results reveal that FINTA has a superior filtering performance compared to state-of-the-art methods. Together, this work brings forward a new deep learning framework in tractography based on autoencoders, and shows how it can be applied for filtering purposes. It sets the foundations for opening up new prospects towards more accurate and robust tractometry and connectivity diffusion MRI analyses, which may ultimately lead to improve the imaging of the white matter anatomy.

Semantic parsing is one of the key components of natural language understanding systems. A successful parse transforms an input utterance to an action that is easily understood by the system. Many algorithms have been proposed to solve this problem, from conventional rulebased or statistical slot-filling systems to shiftreduce based neural parsers. For complex parsing tasks, the state-of-the-art method is based on autoregressive sequence to sequence models to generate the parse directly. This model is slow at inference time, generating parses in O(n) decoding steps (n is the length of the target sequence). In addition, we demonstrate that this method performs poorly in zero-shot cross-lingual transfer learning settings. In this paper, we propose a non-autoregressive parser which is based on the insertion transformer to overcome these two issues. Our approach 1) speeds up decoding by 3x while outperforming the autoregressive model and 2) significantly improves cross-lingual transfer in the low-resource setting by 37% compared to autoregressive baseline. We test our approach on three well-known monolingual datasets: ATIS, SNIPS and TOP. For cross lingual semantic parsing, we use the MultiATIS++ and the multilingual TOP datasets.

Standard methods for differentially private training of deep neural networks replace back-propagated mini-batch gradients with biased and noisy approximations to the gradient. These modifications to training often result in a privacy-preserving model that is significantly less accurate than its non-private counterpart. We hypothesize that alternative training algorithms may be more amenable to differential privacy. Specifically, we examine the suitability of direct feedback alignment (DFA). We propose the first differentially private method for training deep neural networks with DFA and show that it achieves significant gains in accuracy (often by 10-20%) compared to backprop-based differentially private training on a variety of architectures (fully connected, convolutional) and datasets.

Transformer models have garnered a lot of interest in recent years by delivering state-of-the-art performance in a range of Natural Language Processing (NLP) tasks. However, these models can have over a hundred billion parameters, presenting very high computational and memory requirements. We address this challenge through Approximate Computing, specifically targeting the use of Transformers in NLP tasks. Transformers are typically pre-trained and subsequently specialized for specific tasks through transfer learning. Based on the observation that pre-trained Transformers are often over-parameterized for several downstream NLP tasks, we propose a framework to create smaller, faster and in some cases more accurate models. The key cornerstones of the framework are a Significance Analysis (SA) method that identifies components in a pre-trained Transformer that are less significant for a given task, and techniques to approximate the less significant components. Our approximations include pruning of blocks, attention heads and weight groups, quantization of less significant weights and a low-complexity sign-matching based attention mechanism. Our framework can be adapted to produce models that are faster, smaller and/or more accurate, depending on the user's constraints. We apply our framework to seven Transformer models, including optimized models like DistilBERT and Q8BERT, and three downstream tasks. We demonstrate that our framework produces models that are up to 4 faster and up to 14 smaller (with less than 0.5% relative accuracy degradation), or up to 5.5% more accurate with simultaneous improvements of up to 9.83 in model size or 2.94 in speed. Transformer networks with hundreds of billions of parameters, such as T5 ([17]), Megatron ([22]), BERT ([2]), GPT-2 ( [16]) and GPT-3 ([1]), have achieved state-of-the-art performance in several Natural Language Processing tasks. Model sizes are expected to grow further in the future as increasing the number of parameters has been shown to improve performance.

Neural sequence labeling is an important technique employed for many Natural Language Processing (NLP) tasks, such as Named Entity Recognition (NER), slot tagging for dialog systems and semantic parsing. Large-scale pre-trained language models obtain very good performance on these tasks when fine-tuned on large amounts of task-specific labeled data. However, such large-scale labeled datasets are difficult to obtain for several tasks and domains due to the high cost of human annotation as well as privacy and data access constraints for sensitive user applications. This is exacerbated for sequence labeling tasks requiring such annotations at token-level. In this work, we develop techniques to address the label scarcity challenge for neural sequence labeling models. Specifically, we develop self-training and meta-learning techniques for few-shot training of neural sequence taggers, namely MetaST. While self-training serves as an effective mechanism to learn from large amounts of unlabeled data -- meta-learning helps in adaptive sample re-weighting to mitigate error propagation from noisy pseudo-labels. Extensive experiments on six benchmark datasets including two massive multilingual NER datasets and four slot tagging datasets for task-oriented dialog systems demonstrate the effectiveness of our method with around 10% improvement over state-of-the-art systems for the 10-shot setting.

Autoregressive language models pretrained on large corpora have been successful at solving downstream tasks, even with zero-shot usage. However, there is little theoretical justification for their success. This paper considers the following questions: (1) Why should learning the distribution of natural language help with downstream classification tasks? (2) Why do features learned using language modeling help solve downstream tasks with linear classifiers? For (1), we hypothesize, and verify empirically, that classification tasks of interest can be reformulated as next word prediction tasks, thus making language modeling a meaningful pretraining task. For (2), we analyze properties of the cross-entropy objective to show that $\epsilon$-optimal language models in cross-entropy (log-perplexity) learn features that are $\mathcal{O}(\sqrt{\epsilon})$-good on natural linear classification tasks, thus demonstrating mathematically that doing well on language modeling can be beneficial for downstream tasks. We perform experiments to verify assumptions and validate theoretical results. Our theoretical insights motivate a simple alternative to the cross-entropy objective that performs well on some linear classification tasks.