Heeyoul Choi

Richard Kimera, Daniela N. Rim, Heeyoul Choi

Neural machine translation (NMT) has achieved great successes with large datasets, so NMT is more premised on high-resource languages. This continuously underpins the low resource languages such as Luganda due to the lack of high-quality parallel corpora, so even 'Google translate' does not serve Luganda at the time of this writing. In this paper, we build a parallel corpus with 41,070 pairwise sentences for Luganda and English which is based on three different open-sourced corpora. Then, we train NMT models with hyper-parameter search on the dataset. Experiments gave us a BLEU score of 21.28 from Luganda to English and 17.47 from English to Luganda. Some translation examples show high quality of the translation. We believe that our model is the first Luganda-English NMT model. The bilingual dataset we built will be available to the public.

Malintha Fernando, Ransalu Senanayake, Heeyoul Choi et al.

Autonomous mobility is emerging as a new disruptive mode of urban transportation for moving cargo and passengers. However, designing scalable autonomous fleet coordination schemes to accommodate fast-growing mobility systems is challenging primarily due to the increasing heterogeneity of the fleets, time-varying demand patterns, service area expansions, and communication limitations. We introduce the concept of partially observable advanced air mobility games to coordinate a fleet of aerial vehicles by accounting for the heterogeneity of the interacting agents and the self-interested nature inherent to commercial mobility fleets. To model the complex interactions among the agents and the observation uncertainty in the mobility networks, we propose a novel heterogeneous graph attention encoder-decoder (HetGAT Enc-Dec) neural network-based stochastic policy. We train the policy by leveraging deep multi-agent reinforcement learning, allowing decentralized decision-making for the agents using their local observations. Through extensive experimentation, we show that the learned policy generalizes to various fleet compositions, demand patterns, and observation topologies. Further, fleets operating under the HetGAT Enc-Dec policy outperform other state-of-the-art graph neural network policies by achieving the highest fleet reward and fulfillment ratios in on-demand mobility networks.

Namjin Seo, DongNyeong Heo, Heeyoul Choi

Network function virtualization (NFV) and software-defined network (SDN) have become emerging network paradigms, allowing virtualized network function (VNF) deployment at a low cost. Even though VNF deployment can be flexible, it is still challenging to optimize VNF deployment due to its high complexity. Several studies have approached the task as dynamic programming, e.g., integer linear programming (ILP). However, optimizing VNF deployment for highly complex networks remains a challenge. Alternatively, reinforcement learning (RL) based approaches have been proposed to optimize this task, especially to employ a scaling action-based method which can deploy VNFs within less computational time. However, the model architecture can be improved further to generalize to the different networking settings. In this paper, we propose an enhanced model which can be adapted to more general network settings. We adopt the improved GNN architecture and a few techniques to obtain a better node representation for the VNF deployment task. Furthermore, we apply a recently proposed RL method, phasic policy gradient (PPG), to leverage the shared representation of the service function chain (SFC) generation model from the value function. We evaluate the proposed method in various scenarios, achieving a better QoS with minimum resource utilization compared to the previous methods. Finally, as a qualitative evaluation, we analyze our proposed encoder's representation for the nodes, which shows a more disentangled representation.

Hyunsub Lee, Heeyoul Choi

In contrastive learning in the image domain, the anchor and positive samples are forced to have as close representations as possible. However, forcing the two samples to have the same representation could be misleading because the data augmentation techniques make the two samples different. In this paper, we introduce a new representation, partitioned representation, which can learn both common and unique features of the anchor and positive samples in contrastive learning. The partitioned representation consists of two parts: the content part and the style part. The content part represents common features of the class, and the style part represents the own features of each sample, which can lead to the representation of the data augmentation method. We can achieve the partitioned representation simply by decomposing a loss function of contrastive learning into two terms on the two separate representations, respectively. To evaluate our representation with two parts, we take two framework models: Variational AutoEncoder (VAE) and BootstrapYour Own Latent(BYOL) to show the separability of content and style, and to confirm the generalization ability in classification, respectively. Based on the experiments, we show that our approach can separate two types of information in the VAE framework and outperforms the conventional BYOL in linear separability and a few-shot learning task as downstream tasks.

DongNyeong Heo, Daniela Noemi Rim, Heeyoul Choi

Causal language modeling (CLM) serves as the foundational framework underpinning remarkable successes of recent large language models (LLMs). Despite its success, the training approach for next word prediction poses a potential risk of causing the model to overly focus on local dependencies within a sentence. While prior studies have been introduced to predict future N words simultaneously, they were primarily applied to tasks such as masked language modeling (MLM) and neural machine translation (NMT). In this study, we introduce a simple N-gram prediction framework for the CLM task. Moreover, we introduce word difference representation (WDR) as a surrogate and contextualized target representation during model training on the basis of N-gram prediction framework. To further enhance the quality of next word prediction, we propose an ensemble method that incorporates the future N words' prediction results. Empirical evaluations across multiple benchmark datasets encompassing CLM and NMT tasks demonstrate the significant advantages of our proposed methods over the conventional CLM.

DongNyeong Heo, Heeyoul Choi

Language models (LMs) are a central component of modern AI systems, and diffusion-based language models (DLMs) have recently emerged as a competitive alternative. Both paradigms rely on word embeddings not only to represent the input sentence, but also to represent the target sentence that backbone models are trained to predict. We argue that such static embedding of the target word is insensitive to neighboring words, encouraging locally accurate word prediction while neglecting global structure across the target sentence. To address this limitation, we propose a continuous sentence representation, termed sentence curve, defined as a spline curve whose control points affect multiple words in the sentence. Based on this representation, we introduce sentence curve language model (SCLM), which extends DLMs to predict sentence curves instead of the static word embeddings. We theoretically show that sentence curve prediction induces a regularization effect that promotes global structure modeling, and characterize how different sentence curve types affect this behavior. Empirically, SCLM achieves SOTA performance among DLMs on IWSLT14 and WMT14, shows stable training without burdensome knowledge distillation, and demonstrates promising potential compared to discrete DLMs on LM1B.

Harim Kim, Yuhan Wang, Minkyu Ahn et al.

Unsupervised anomaly detection (UAD) in medical imaging is crucial for identifying pathological abnormalities without requiring extensive labeled data. However, existing diffusion-based UAD models rely solely on imaging features, limiting their ability to distinguish between normal anatomical variations and pathological anomalies. To address this, we propose Diff3M, a multi-modal diffusion-based framework that integrates chest X-rays and structured Electronic Health Records (EHRs) for enhanced anomaly detection. Specifically, we introduce a novel image-EHR cross-attention module to incorporate structured clinical context into the image generation process, improving the model's ability to differentiate normal from abnormal features. Additionally, we develop a static masking strategy to enhance the reconstruction of normal-like images from anomalies. Extensive evaluations on CheXpert and MIMIC-CXR/IV demonstrate that Diff3M achieves state-of-the-art performance, outperforming existing UAD methods in medical imaging. Our code is available at this http URL https://github.com/nth221/Diff3M

DongNyeong Heo, Heeyoul Choi

Back-translation (BT) is an effective semi-supervised learning framework in neural machine translation (NMT). A pre-trained NMT model translates monolingual sentences and makes synthetic bilingual sentence pairs for the training of the other NMT model, and vice versa. Understanding the two NMT models as inference and generation models, respectively, the training method of variational auto-encoder (VAE) was applied in previous works, which is a mainstream framework of generative models. However, the discrete property of translated sentences prevents gradient information from flowing between the two NMT models. In this paper, we propose the categorical reparameterization trick (CRT) that makes NMT models generate differentiable sentences so that the VAE's training framework can work in an end-to-end fashion. Our BT experiment conducted on a WMT benchmark dataset demonstrates the superiority of our proposed CRT compared to the Gumbel-softmax trick, which is a popular reparameterization method for categorical variable. Moreover, our experiments conducted on multiple WMT benchmark datasets demonstrate that our proposed end-to-end training framework is effective in terms of BLEU scores not only compared to its counterpart baseline which is not trained in an end-to-end fashion, but also compared to other previous BT works. The code is available at the web.

Daniela N. Rim, Heeyoul Choi

Traditional Convolutional Neural Networks have been successful in capturing local, position-invariant features in text, but their capacity to model complex transformation within language can be further explored. In this work, we explore a novel approach by integrating Lie Convolutions into Convolutional-based sentence classifiers, inspired by the ability of Lie group operations to capture complex, non-Euclidean symmetries. Our proposed models SCLie and DPCLie empirically outperform traditional Convolutional-based sentence classifiers, suggesting that Lie-based models relatively improve the accuracy by capturing transformations not commonly associated with language. Our findings motivate more exploration of new paradigms in language modeling.

Daniela N. Rim, Heeyoul Choi

This work explores the capability of conversational chatbots powered by large language models (LLMs), to understand and characterize predicate symmetry, a cognitive linguistic function traditionally believed to be an inherent human trait. Leveraging in-context learning (ICL), a paradigm shift enabling chatbots to learn new tasks from prompts without re-training, we assess the symmetrical reasoning of five chatbots: ChatGPT 4, Huggingface chat AI, Microsoft's Copilot AI, LLaMA through Perplexity, and Gemini Advanced. Using the Symmetry Inference Sentence (SIS) dataset by Tanchip et al. (2020), we compare chatbot responses against human evaluations to gauge their understanding of predicate symmetry. Experiment results reveal varied performance among chatbots, with some approaching human-like reasoning capabilities. Gemini, for example, reaches a correlation of 0.85 with human scores, while providing a sounding justification for each symmetry evaluation. This study underscores the potential and limitations of LLMs in mirroring complex cognitive processes as symmetrical reasoning.

Daniela N. Rim, Kimera Richard, Heeyoul Choi

The Transformer model has revolutionized Natural Language Processing tasks such as Neural Machine Translation, and many efforts have been made to study the Transformer architecture, which increased its efficiency and accuracy. One potential area for improvement is to address the computation of empty tokens that the Transformer computes only to discard them later, leading to an unnecessary computational burden. To tackle this, we propose an algorithm that sorts translation sentence pairs based on their length before batching, minimizing the waste of computing power. Since the amount of sorting could violate the independent and identically distributed (i.i.d) data assumption, we sort the data partially. In experiments, we apply the proposed method to English-Korean and English-Luganda language pairs for machine translation and show that there are gains in computational time while maintaining the performance. Our method is independent of architectures, so that it can be easily integrated into any training process with flexible data lengths.

DongNyeong Heo, Heeyoul Choi

The Transformer architecture has become widely adopted due to its demonstrated success, attributed to the attention mechanism at its core. Despite these successes, the attention mechanism of Transformers is associated with two well-known issues: rank-collapse and gradient vanishing. In this paper, we present a theoretical analysis that it is inherently difficult to address both issues simultaneously in the conventional attention mechanism. To handle these issues, we introduce a novel class of attention mechanism, referred to as generalized probabilistic attention mechanism (GPAM), and its dual-attention implementation within the Transformer architecture. Unlike conventional attention mechanisms, GPAM allows for negative attention scores while preserving a fixed total sum. We provide theoretical evidence that the proposed dual-attention GPAM (daGPAM) effectively mitigates both the rank-collapse and gradient vanishing issues which are difficult to resolve simultaneously with the conventional attention mechanisms. Furthermore, we empirically validate this theoretical evidence, demonstrating the superiority of daGPAM compared to other alternative attention mechanisms that were proposed to address the same issues. Additionally, we demonstrate the practical benefits of GPAM in natural language processing tasks, such as language modeling and neural machine translation.

Richard Kimera, Dongnyeong Heo, Daniela N. Rim et al.

In this paper,we explore the application of Back translation (BT) as a semi-supervised technique to enhance Neural Machine Translation(NMT) models for the English-Luganda language pair, specifically addressing the challenges faced by low-resource languages. The purpose of our study is to demonstrate how BT can mitigate the scarcity of bilingual data by generating synthetic data from monolingual corpora. Our methodology involves developing custom NMT models using both publicly available and web-crawled data, and applying Iterative and Incremental Back translation techniques. We strategically select datasets for incremental back translation across multiple small datasets, which is a novel element of our approach. The results of our study show significant improvements, with translation performance for the English-Luganda pair exceeding previous benchmarks by more than 10 BLEU score units across all translation directions. Additionally, our evaluation incorporates comprehensive assessment metrics such as SacreBLEU, ChrF2, and TER, providing a nuanced understanding of translation quality. The conclusion drawn from our research confirms the efficacy of BT when strategically curated datasets are utilized, establishing new performance benchmarks and demonstrating the potential of BT in enhancing NMT models for low-resource languages.

Richard Kimera, Yun-Seon Kim, Heeyoul Choi

This paper addresses the ethical challenges of Artificial Intelligence in Neural Machine Translation (NMT) systems, emphasizing the imperative for developers to ensure fairness and cultural sensitivity. We investigate the ethical competence of AI models in NMT, examining the Ethical considerations at each stage of NMT development, including data handling, privacy, data ownership, and consent. We identify and address ethical issues through empirical studies. These include employing Transformer models for Luganda-English translations and enhancing efficiency with sentence mini-batching. And complementary studies that refine data labeling techniques and fine-tune BERT and Longformer models for analyzing Luganda and English social media content. Our second approach is a literature review from databases such as Google Scholar and platforms like GitHub. Additionally, the paper probes the distribution of responsibility between AI systems and humans, underscoring the essential role of human oversight in upholding NMT ethical standards. Incorporating a biblical perspective, we discuss the societal impact of NMT and the broader ethical responsibilities of developers, positing them as stewards accountable for the societal repercussions of their creations.

DongNyeong Heo, Daniela Noemi Rim, Heeyoul Choi

An internet network service provider manages its network with multiple objectives, such as high quality of service (QoS) and minimum computing resource usage. To achieve these objectives, a reinforcement learning-based (RL) algorithm has been proposed to train its network management agent. Usually, their algorithms optimize their agents with respect to a single static reward formulation consisting of multiple objectives with fixed importance factors, which we call preferences. However, in practice, the preference could vary according to network status, external concerns and so on. For example, when a server shuts down and it can cause other servers' traffic overloads leading to additional shutdowns, it is plausible to reduce the preference of QoS while increasing the preference of minimum computing resource usages. In this paper, we propose new RL-based network management agents that can select actions based on both states and preferences. With our proposed approach, we expect a single agent to generalize on various states and preferences. Furthermore, we propose a numerical method that can estimate the distribution of preference that is advantageous for unbiased training. Our experiment results show that the RL agents trained based on our proposed approach significantly generalize better with various preferences than the previous RL approaches, which assume static preference during training. Moreover, we demonstrate several analyses that show the advantages of our numerical estimation method.

Richard Kimera, Daniela N. Rim, Joseph Kirabira et al.

Depression is a global burden and one of the most challenging mental health conditions to control. Experts can detect its severity early using the Beck Depression Inventory (BDI) questionnaire, administer appropriate medication to patients, and impede its progression. Due to the fear of potential stigmatization, many patients turn to social media platforms like Reddit for advice and assistance at various stages of their journey. This research extracts text from Reddit to facilitate the diagnostic process. It employs a proposed labeling approach to categorize the text and subsequently fine-tunes the Longformer model. The model's performance is compared against baseline models, including Naive Bayes, Random Forest, Support Vector Machines, and Gradient Boosting. Our findings reveal that the Longformer model outperforms the baseline models in both English (48%) and Luganda (45%) languages on a custom-made dataset.

DongNyeong Heo, Heeyoul Choi

Non-autoregressive neural machine translation (NAT) offers substantial translation speed up compared to autoregressive neural machine translation (AT) at the cost of translation quality. Latent variable modeling has emerged as a promising approach to bridge this quality gap, particularly for addressing the chronic multimodality problem in NAT. In the previous works that used latent variable modeling, they added an auxiliary model to estimate the posterior distribution of the latent variable conditioned on the source and target sentences. However, it causes several disadvantages, such as redundant information extraction in the latent variable, increasing the number of parameters, and a tendency to ignore some information from the inputs. In this paper, we propose a novel latent variable modeling that integrates a dual reconstruction perspective and an advanced hierarchical latent modeling with a shared intermediate latent space across languages. This latent variable modeling hypothetically alleviates or prevents the above disadvantages. In our experiment results, we present comprehensive demonstrations that our proposed approach infers superior latent variables which lead better translation quality. Finally, in the benchmark translation tasks, such as WMT, we demonstrate that our proposed method significantly improves translation quality compared to previous NAT baselines including the state-of-the-art NAT model.

Jin Ha Woo, Heeyoul Choi

The Second Language Acquisition field has been significantly impacted by a greater emphasis on individualized learning and rapid developments in artificial intelligence (AI). Although increasingly adaptive language learning tools are being developed with the application of AI to the Computer Assisted Language Learning field, there have been concerns regarding insufficient information and teacher preparation. To effectively utilize these tools, teachers need an in-depth overview on recently developed AI-based language learning tools. Therefore, this review synthesized information on AI tools that were developed between 2017 and 2020. A majority of these tools utilized machine learning and natural language processing, and were used to identify errors, provide feedback, and assess language abilities. After using these tools, learners demonstrated gains in their language abilities and knowledge. This review concludes by presenting pedagogical implications and emerging themes in the future research of AI-based language learning tools.

Chungjun Lee, Jibum Hong, DongNyeong Heo et al.

Software-defined networking (SDN) and network function virtualization (NFV) have enabled the efficient provision of network service. However, they also raised new tasks to monitor and ensure the status of virtualized service, and anomaly detection is one of such tasks. There have been many data-driven approaches to implement anomaly detection system (ADS) for virtual network functions in service function chains (SFCs). In this paper, we aim to develop more advanced deep learning models for ADS. Previous approaches used learning algorithms such as random forest (RF), gradient boosting machine (GBM), or deep neural networks (DNNs). However, these models have not utilized sequential dependencies in the data. Furthermore, they are limited as they can only apply to the SFC setting from which they were trained. Therefore, we propose several sequential deep learning models to learn time-series patterns and sequential patterns of the virtual network functions (VNFs) in the chain with variable lengths. As a result, the suggested models improve detection performance and apply to SFCs with varying numbers of VNFs.

Daniela N. Rim, DongNyeong Heo, Heeyoul Choi

For years, adversarial training has been extensively studied in natural language processing (NLP) settings. The main goal is to make models robust so that similar inputs derive in semantically similar outcomes, which is not a trivial problem since there is no objective measure of semantic similarity in language. Previous works use an external pre-trained NLP model to tackle this challenge, introducing an extra training stage with huge memory consumption during training. However, the recent popular approach of contrastive learning in language processing hints a convenient way of obtaining such similarity restrictions. The main advantage of the contrastive learning approach is that it aims for similar data points to be mapped close to each other and further from different ones in the representation space. In this work, we propose adversarial training with contrastive learning (ATCL) to adversarially train a language processing task using the benefits of contrastive learning. The core idea is to make linear perturbations in the embedding space of the input via fast gradient methods (FGM) and train the model to keep the original and perturbed representations close via contrastive learning. In NLP experiments, we applied ATCL to language modeling and neural machine translation tasks. The results show not only an improvement in the quantitative (perplexity and BLEU) scores when compared to the baselines, but ATCL also achieves good qualitative results in the semantic level for both tasks without using a pre-trained model.

Daniela N. Rim, Inseon Jang, Heeyoul Choi

Recent achievements in end-to-end deep learning have encouraged the exploration of tasks dealing with highly structured data with unified deep network models. Having such models for compressing audio signals has been challenging since it requires discrete representations that are not easy to train with end-to-end backpropagation. In this paper, we present an end-to-end deep learning approach that combines recurrent neural networks (RNNs) within the training strategy of variational autoencoders (VAEs) with a binary representation of the latent space. We apply a reparametrization trick for the Bernoulli distribution for the discrete representations, which allows smooth backpropagation. In addition, our approach allows the separation of the encoder and decoder, which is necessary for compression tasks. To our best knowledge, this is the first end-to-end learning for a single audio compression model with RNNs, and our model achieves a Signal to Distortion Ratio (SDR) of 20.54.

DongNyeong Heo, Doyoung Lee, Hee-Gon Kim et al.

In the management of computer network systems, the service function chaining (SFC) modules play an important role by generating efficient paths for network traffic through physical servers with virtualized network functions (VNF). To provide the highest quality of services, the SFC module should generate a valid path quickly even in various network topology situations including dynamic VNF resources, various requests, and changes of topologies. The previous supervised learning method demonstrated that the network features can be represented by graph neural networks (GNNs) for the SFC task. However, the performance was limited to only the fixed topology with labeled data. In this paper, we apply reinforcement learning methods for training models on various network topologies with unlabeled data. In the experiments, compared to the previous supervised learning method, the proposed methods demonstrated remarkable flexibility in new topologies without re-designing and re-training, while preserving a similar level of performance.

DongNyeong Heo, Stanislav Lange, Hee-Gon Kim et al.

Software-defined networking (SDN) and the network function virtualization (NFV) led to great developments in software based control technology by decreasing expenditures. Service function chaining (SFC) is an important technology to find efficient paths in network servers to process all of the requested virtualized network functions (VNF). However, SFC is challenging since it has to maintain high Quality of Service (QoS) even for complicated situations. Although some works have been conducted for such tasks with high-level intelligent models like deep neural networks (DNNs), those approaches are not efficient in utilizing the topology information of networks and cannot be applied to networks with dynamically changing topology since their models assume that the topology is fixed. In this paper, we propose a new neural network architecture for SFC, which is based on graph neural network (GNN) considering the graph-structured properties of network topology. The proposed SFC model consists of an encoder and a decoder, where the encoder finds the representation of the network topology, and then the decoder estimates probabilities of neighborhood nodes and their probabilities to process a VNF. In the experiments, our proposed architecture outperformed previous performances of DNN based baseline model. Moreover, the GNN based model can be applied to a new network topology without re-designing and re-training.

Hayoung Um, Heeyoul Choi

In pattern recognition or machine learning, it is a very fundamental task to find nearest neighbors of a given point. All the methods for the task work basically by comparing the given point to all the points in the data set. That is why the computational cost increases with the number of data points. However, the human visual system seems to work in a different way. When the human visual system tries to find the neighbors of one point on a map, it directly focuses on the area around the point and actively searches the neighbors by looking or zooming in and out around the point. In this paper, we propose an innovative search method for nearest neighbors, which seems very similar to how human visual system works on the task.

Hyeokmin Gwon, Chungjun Lee, Rakun Keum et al.

Growing number of network devices and services have led to increasing demand for protective measures as hackers launch attacks to paralyze or steal information from victim systems. Intrusion Detection System (IDS) is one of the essential elements of network perimeter security which detects the attacks by inspecting network traffic packets or operating system logs. While existing works demonstrated effectiveness of various machine learning techniques, only few of them utilized the time-series information of network traffic data. Also, categorical information has not been included in neural network based approaches. In this paper, we propose network intrusion detection models based on sequential information using long short-term memory (LSTM) network and categorical information using the embedding technique. We have experimented the models with UNSW-NB15, which is a comprehensive network traffic dataset. The experiment results confirm that the proposed method improve the performance, observing binary classification accuracy of 99.72\%.

Sangchul Hahn, Heeyoul Choi

Since deep learning became a key player in natural language processing (NLP), many deep learning models have been showing remarkable performances in a variety of NLP tasks, and in some cases, they are even outperforming humans. Such high performance can be explained by efficient knowledge representation of deep learning models. While many methods have been proposed to learn more efficient representation, knowledge distillation from pretrained deep networks suggest that we can use more information from the soft target probability to train other neural networks. In this paper, we propose a new knowledge distillation method self-knowledge distillation, based on the soft target probabilities of the training model itself, where multimode information is distilled from the word embedding space right below the softmax layer. Due to the time complexity, our method approximates the soft target probabilities. In experiments, we applied the proposed method to two different and fundamental NLP tasks: language model and neural machine translation. The experiment results show that our proposed method improves performance on the tasks.

Sangchul Hahn, Heeyoul Choi

After deep generative models were successfully applied to image generation tasks, learning disentangled latent variables of data has become a crucial part of deep generative model research. Many models have been proposed to learn an interpretable and factorized representation of latent variable by modifying their objective function or model architecture. To disentangle the latent variable, some models show lower quality of reconstructed images and others increase the model complexity which is hard to train. In this paper, we propose a simple disentangling method based on a traditional whitening process. The proposed method is applied to the latent variables of variational auto-encoder (VAE), although it can be applied to any generative models with latent variables. In experiment, we apply the proposed method to simple VAE models and experiment results confirm that our method finds more interpretable factors from the latent space while keeping the reconstruction error the same as the conventional VAE's error.

Hayoung Eom, Heeyoul Choi

Convolutional neural networks (CNNs) have achieved remarkable performance in many applications, especially in image recognition tasks. As a crucial component of CNNs, sub-sampling plays an important role for efficient training or invariance property, and max-pooling and arithmetic average-pooling are commonly used sub-sampling methods. In addition to the two pooling methods, however, there could be many other pooling types, such as geometric average, harmonic average, and so on. Since it is not easy for algorithms to find the best pooling method, usually the pooling types are assumed a priority, which might not be optimal for different tasks. In line with the deep learning philosophy, the type of pooling can be driven by data for a given task. In this paper, we propose {\it $α$-integration pooling} ($α$I-pooling), which has a trainable parameter $α$ to find the type of pooling. $α$I-pooling is a general pooling method including max-pooling and arithmetic average-pooling as a special case, depending on the parameter $α$. Experiments show that $α$I-pooling outperforms other pooling methods including max-pooling, in image recognition tasks. Also, it turns out that each layer has different optimal pooling type.

Sangchul Hahn, Heeyoul Choi

As one of standard approaches to train deep neural networks, dropout has been applied to regularize large models to avoid overfitting, and the improvement in performance by dropout has been explained as avoiding co-adaptation between nodes. However, when correlations between nodes are compared after training the networks with or without dropout, one question arises if co-adaptation avoidance explains the dropout effect completely. In this paper, we propose an additional explanation of why dropout works and propose a new technique to design better activation functions. First, we show that dropout can be explained as an optimization technique to push the input towards the saturation area of nonlinear activation function by accelerating gradient information flowing even in the saturation area in backpropagation. Based on this explanation, we propose a new technique for activation functions, {\em gradient acceleration in activation function (GAAF)}, that accelerates gradients to flow even in the saturation area. Then, input to the activation function can climb onto the saturation area which makes the network more robust because the model converges on a flat region. Experiment results support our explanation of dropout and confirm that the proposed GAAF technique improves image classification performance with expected properties.

Heeyoul Choi

Recurrent neural networks (RNNs) have been drawing much attention with great success in many applications like speech recognition and neural machine translation. Long short-term memory (LSTM) is one of the most popular RNN units in deep learning applications. LSTM transforms the input and the previous hidden states to the next states with the affine transformation, multiplication operations and a nonlinear activation function, which makes a good data representation for a given task. The affine transformation includes rotation and reflection, which change the semantic or syntactic information of dimensions in the hidden states. However, considering that a model interprets the output sequence of LSTM over the whole input sequence, the dimensions of the states need to keep the same type of semantic or syntactic information regardless of the location in the sequence. In this paper, we propose a simple variant of the LSTM unit, persistent recurrent unit (PRU), where each dimension of hidden states keeps persistent information across time, so that the space keeps the same meaning over the whole sequence. In addition, to improve the nonlinear transformation power, we add a feedforward layer in the PRU structure. In the experiment, we evaluate our proposed methods with three different tasks, and the results confirm that our methods have better performance than the conventional LSTM.

Heeyoul Choi, Kyunghyun Cho, Yoshua Bengio

Neural machine translation (NMT) has been a new paradigm in machine translation, and the attention mechanism has become the dominant approach with the state-of-the-art records in many language pairs. While there are variants of the attention mechanism, all of them use only temporal attention where one scalar value is assigned to one context vector corresponding to a source word. In this paper, we propose a fine-grained (or 2D) attention mechanism where each dimension of a context vector will receive a separate attention score. In experiments with the task of En-De and En-Fi translation, the fine-grained attention method improves the translation quality in terms of BLEU score. In addition, our alignment analysis reveals how the fine-grained attention mechanism exploits the internal structure of context vectors.

Sungjin Ahn, Heeyoul Choi, Tanel Pärnamaa et al.

Current language models have a significant limitation in the ability to encode and decode factual knowledge. This is mainly because they acquire such knowledge from statistical co-occurrences although most of the knowledge words are rarely observed. In this paper, we propose a Neural Knowledge Language Model (NKLM) which combines symbolic knowledge provided by the knowledge graph with the RNN language model. By predicting whether the word to generate has an underlying fact or not, the model can generate such knowledge-related words by copying from the description of the predicted fact. In experiments, we show that the NKLM significantly improves the performance while generating a much smaller number of unknown words.

Heeyoul Choi, Kyunghyun Cho, Yoshua Bengio

We first observe a potential weakness of continuous vector representations of symbols in neural machine translation. That is, the continuous vector representation, or a word embedding vector, of a symbol encodes multiple dimensions of similarity, equivalent to encoding more than one meaning of the word. This has the consequence that the encoder and decoder recurrent networks in neural machine translation need to spend substantial amount of their capacity in disambiguating source and target words based on the context which is defined by a source sentence. Based on this observation, in this paper we propose to contextualize the word embedding vectors using a nonlinear bag-of-words representation of the source sentence. Additionally, we propose to represent special tokens (such as numbers, proper nouns and acronyms) with typed symbols to facilitate translating those words that are not well-suited to be translated via continuous vectors. The experiments on En-Fr and En-De reveal that the proposed approaches of contextualization and symbolization improves the translation quality of neural machine translation systems significantly.