In this relay series, Samsung Newsroom is introducing tech experts from Samsung’s R&D centers around the globe to hear more about the work they do and the ways in which it is directly improving the lives of consumers.

The third expert in the series to be introduced is Bin Dai, Staff Engineer at the Artificial Intelligence (AI) Lab in Samsung R&D Institute China – Beijing (SRC-B). Dai joined SRC-B in 2020 to join his colleagues in working on network compression and on-device model design and research. Read on to learn more about the groundbreaking technologies Dai and his team are developing at SRC-B.

Q: AI-based technologies, including NLP (Natural Language Processing) and acoustic intelligence, are cutting-edge research areas that are constantly breaking new ground. But what role does the core research offering provided by machine learning play as a background for these innovations?

Machine learning plays a crucial role into bringing all kinds of technologies directly to users. Computer vision and speech recognition are two of the most successful areas currently utilizing AI. However, existing AI algorithms require huge computation resources, making it difficult to deploy state-of-the-art algorithms on mobile devices. In order to fix this issue, our AI Lab is working on producing tiny models with powerful performance from both a theoretical and a practical perspective. In this way, our core research is set to innovate all kinds of AI-based technologies.

Q: Can you please briefly introduce the Beijing Research Institute, and the kind of work that goes on there?

SRC-B is one of Samsung’s Electronics’ advanced R&D centers and was established in 2000, the first Samsung R&D center to be established in China. SRC-B focuses on groundbreaking technologies and specializes in artificial intelligence (AI) and next-generation telecommunications, from machine learning, computer vision, language processing and voice intelligence through to 3GPP standardization and more. We also promote tight industrial-academic partnerships. In April 2019, the AI Lab was established to focus on fundamental research into machine learning, and we are continuously looking for ways to apply our research results to Samsung products.

Q: Following the success of your major research thesis and other accomplishments, what are you working on at the moment?

SRC-B is currently aiming to find the best possible way to enhance the accuracy of an AI algorithm while reducing the computation complexity and resources used to do so. In order to achieve these goals, we are currently working on two research topics that enable accurate predictions with less data required: equivariant networks, part of the broader topic of geometric deep learning, and dynamic inference. There are many kinds of symmetries in computer vision datasets which are able to provide accurate depth measurement like human eyes can, such as image and LiDAR point clouds. With an equivariant network, these symmetries are taken into consideration when designing the network. It is thus able to achieve better performance with fewer resources since we have specifically considered the intrinsic structure of the dataset.

Dynamic inference is also a very interesting research direction. Unlike conventional methods which harness a fixed architecture for all data samples, dynamic inference can adaptively decide how many resources to use for each data sample. Accordingly, it will use fewer computational resources for simple samples and more resources for difficult ones. By doing so, the average computation resource used can be significantly reduced.

Q: Fundamental research into AI has been empowering all kinds of user-forward application fields, from computer vision to speech recognition. Could you explain a bit more about why this is, and the direction of research you and the AI Lab have been taking in order to optimize mobile experiences?

In this era of the internet, data is flooding everywhere around us. Where there is data, there is knowledge. AI algorithms are the very best tool for uncovering the knowledge hidden behind the data and make use of this knowledge to make all of our lives better.

We have developed a network compression algorithm based on the information bottleneck theory – which posits that extraneous details can be removed from noisy input data as if squeezed through a bottleneck – which has been applied to multiple tasks including video recognition, image segmentation and machine translation. We also actively collaborate with other labs in SRC-B in order to develop more powerful AI algorithms, including the Neural Architecture Search (NAS) and Once-For-All (OFA) solutions.

Q: What do you see as the main user benefits from incorporating all base mobile technologies with machine learning-based AI technologies?

Machine learning-based AI technologies can dramatically improve users’ lives in three key ways. Firstly, there are many convenient functions that simply cannot work without AI technologies. For example, the automatic question and answering system on mobile devices has to be powered by AI algorithms. Other more traditional methods are only able to handle very limited, pre-defined questions.

Secondly, AI techniques can significantly improve the performance of many applications compared to their performance when harnessing conventional technologies only. For example, after applying deep neural networks to a camera’s neural image signal processing (ISP) function, the quality of photos taken on that camera becomes significantly better.

Thirdly, AI technologies are capable of providing services that users previously didn’t even know they needed. For example, AI is capable of developing a user-specific software based on that user’s specific preferences, meaning that the user’s device experience can continuously be improved.

▲ Researchers at Samsung R&D Institute China – Beijing

Q: How does the work you do synergize with the work undertaken by the rest of Samsung R&D Institute China – Beijing, or perhaps even other R&D Institutes around the world? How does it come together to make users’ lives more convenient?

We are constantly collaborating with the other teams within SRC-B. We have been collaborating recently with our Visual Computing team in order to apply our information bottleneck-based compression algorithm to video recognition tasks and human segmentation tasks, resulting in the significant reduction of model sizes without any performance drop. In 2021, we participated in the Conference on Computer Vision and Pattern Recognition (CVPR)’s Neural Architecture Search (NAS) competition as one team with this solution, and won 1^st place.

We have also been working with our Language Intelligence team to compress their machine translation model, which facilitates the commercialization of their application.

We also believe that we can produce better research and application results by further communication, discussion and collaboration with AI centers globally.

Q: What do you see as being the main trends within your industry right now? How have you been incorporating them into the research you do at Samsung R&D Institute China – Beijing?

There are a lot of trending topics within our field at this time. Efficient network architecture design, self-supervised learning and graph neural networks are just a few examples.

Our focus is on network compression and tiny model design, which is ultimately useful for applications on mobile devices. There are a lot of mobile devices, such as smartphones, that possess very limited computational resources, meaning that it is impossible to deploy the huge models designed for services to these devices. Therefore, my team is focused on designing models suitable for these devices.

There are different ways to achieve these kinds of light yet powerful models. For instance, network pruning, quantization, knowledge distillation, neural network architecture search and dynamic inference are just a few industry areas that we are focusing on right now to achieve this.

Q: What has been the achievement at Samsung R&D Institute China – Beijing that you are most proud of so far?

Developed together in collaboration with our Communication Research team, we engineered AI algorithms for wireless communication. This solution achieved first place at the Wireless Communication AI Competition (WAIC) this year, which is the official competition for 5G+AI in China with over 600 teams enter from around the world and is held by the China Academy of Information and Communication Technology (CAICT). I am proud of this achievement and feel that it validates my belief that 5G combined with AI is a research direction with great potential.

An interview with Evgeny Pavlov, a system software expert from Samsung R&D Institute Russia (SRR) can be found in the following episode.

Spearheading Speech Translation Research

Samsung R&D Institute Poland and Samsung R&D Institute China-Beijing competed with some of the world’s top universities and research labs to win first place in two separate challenges at the International Workshop on Spoken Language Translation (IWSLT), one of the world’s longest-running workshops on automatic language translation. This year, IWSLT joined the Association for Computational Linguistics conference (ACL), a premier conference in the field of computational linguistics, to cover a broad spectrum of research areas that are concerned with computational approaches to natural language.

For the Offline Speech Translation task, which assesses the translation of TED talks from English to German, Samsung R&D Institute Poland won first place for the second time with its own research capabilities in audio to text translation. The conferral of this award marks the fourth consecutive year that teams from Samsung R&D Institute Poland have taken first prize in IWSLT challenges, including previous years’ text translation tasks.

This year’s Offline Speech Translation task allowed participants to submit systems based on either the traditional speech translation pipeline system composed of an automatic speech recognition (ASR) and a machine translation (MT) or an End-to-End (E2E) system. Samsung R&D Institute Poland’s system is based on a single encoder-decoder deep neural network – an E2E system – capable of both English and German texts.

In computational linguistics, E2E systems are harnessed to solve the common problem of error accumulation, wherein, in a traditional pipeline, an error in the speech recognition phase can lead to a nonsensical translation. However, research from over the past three years has shown that traditional systems have constantly been outperforming E2E speech translation systems. The Samsung team’s system not only placed first in the E2E category, but also outscored all traditional pipeline system entrants, a remarkable achievement that puts Samsung R&D Institute Poland at the forefront of speech translation research.

The team from Samsung R&D Institute Poland Team who participated in this year’s IWSLT challenges

Innovative Approaches in the Field of Computational Linguistics AI

Samsung R&D Institute China-Beijing took part in a second challenge, the Open Domain Translation task evaluating Japanese to Chinese translation capability, ultimately taking first place. The main goals of this task were the promotion of research into translation between Asian language, the exploitation of noisy parallel web corpora for machine translation and the thoughtful handling of data provenance.

Samsung R&D Institute China-Beijing submitted a system based on Transformer model architecture and adopted the relative position attention. The team focused on improving the Transform baseline system with elaborate data preprocessing and managed to achieve significant improvements. The team also tried shared and exclusive word embedding and compared different granularity of tokens, approaching the process at a sub-word level, including Byte Pair Encoding (BPE) and Sentence Piece. Large-scale back translation on monolingual corpus was used to improve the Neural Machine Translation (NMT) performance.

Members of the team from Samsung R&D Institute China-Beijing Team who participated in this year’s IWSLT challenges

Achievements in AI Audio Signal Interpretation

In addition to their first-place finish in the IWSLT challenge, Samsung R&D Institute Poland was also recognized as one of the leading teams at the Detection and Classification of Acoustic Scenes and Events (DCASE) 2020 challenge, held by IEEE (Institute of Electrical and Electronics Engineers), which aims to use state-of-the-art AI technology to understand and interpret audio signals.

Engineers from Samsung R&D Institute Poland, who possess previous experience in Acoustic Scene Understanding and Sound Sources Localization tasks (having ranked first place in two tasks in 2019), set their focus on Task 2: Unsupervised Detection of Anomalous Sounds for Machine Condition Monitoring. The goal of this task was to identify whether the sound emitted from a target machine was normal or anomalous. The main challenge was detecting unknown anomalous sounds under a condition within which only normal sound samples have been provided as training data. The engineers scored second place out of 40 teams.

The team from Samsung R&D Institute Poland Team who participated in this year’s DCASE challenge

Envisaging the Future of Computer Vision and Pattern Recognition

In June, Samsung R&D Institute China-Beijing also participated in three challenges hosted by the 2020 Conference on Computer Vision and Pattern Recognition (CVPR 2020): the Embodied AI Challenge, the VizWiz-Captions Challenge and the VATEX Video Captioning Challenge. The team claimed second place in the challenges.

The Embodied AI Challenge aimed to enable robots to understand human commands and perform correct actions within a virtual environment, while the VizWiz-Captions Challenge involved predicting an accurate caption when given an image taken by a visually impaired person and the VATEX Video Captioning Challenge aimed to benchmark progress towards models that can describe videos in various languages including English and Chinese.

Members of the team from Samsung R&D Institute China-Beijing Team who participated in this year’s CVPR challenges