Andrew Shin

Andrew Shin

We present AI-Gram, a live platform enabling image-based interactions, to study social dynamics in a fully autonomous multi-agent visual network where all participants are LLM-driven agents. Using the platform, we conduct experiments on how agents communicate and adapt through visual media, and observe the spontaneous emergence of visual reply chains, indicating rich communicative structure. At the same time, agents exhibit aesthetic sovereignty resisting stylistic convergence toward social partners, anchoring under adversarial influence, and a decoupling between visual similarity and social ties. These results reveal a fundamental asymmetry in current agent architectures: strong expressive communication paired with a steadfast preservation of individual visual identity. We release AI-Gram as a publicly accessible, continuously evolving platform for studying social dynamics in Al-native multi-agent systems. https://ai-gram.ai/

Andrew Shin

While large language models (LLMs) have achieved remarkable performance in various tasks including mathematical reasoning, their development typically demands prohibitive computational resources. Recent advancements have reduced costs for training capable models, yet even these approaches rely on high-end hardware clusters. In this paper, we demonstrate that a single average gaming GPU can train a solid mathematical reasoning model, by integrating reinforcement learning and memory optimization techniques. Specifically, we train a 1.5B parameter mathematical reasoning model on RTX 3080 Ti of 16GB memory that achieves comparable or better performance on mathematical reasoning benchmarks than models several times larger, in resource-constrained environments. Our results challenge the paradigm that state-of-the-art mathematical reasoning necessitates massive infrastructure, democratizing access to high-performance AI research. https://github.com/shinandrew/YouronMath.

Andrew Shin, Kunitake Kaneko

Large language models (LLMs) have demonstrated remarkable performances on a wide range of natural language tasks. Yet, LLMs' successes have been largely restricted to tasks concerning words, sentences, or documents, and it remains questionable how much they understand the minimal units of text, namely characters. In this paper, we examine contemporary LLMs regarding their ability to understand character composition of words, and show that most of them fail to reliably carry out even the simple tasks that can be handled by humans with perfection. We analyze their behaviors with comparison to token level performances, and discuss the potential directions for future research.

Andrew Shin, Yusuke Mori, Kunitake Kaneko

Text-to-video generation task has witnessed a notable progress, with the generated outcomes reflecting the text prompts with high fidelity and impressive visual qualities. However, current text-to-video generation models are invariably focused on conveying the visual elements of a single scene, and have so far been indifferent to another important potential of the medium, namely a storytelling. In this paper, we examine text-to-video generation from a storytelling perspective, which has been hardly investigated, and make empirical remarks that spotlight the limitations of current text-to-video generation scheme. We also propose an evaluation framework for storytelling aspects of videos, and discuss the potential future directions.

Andrew Shin, Kunitake Kaneko

Large language models (LLMs) excel at modeling relationships between strings in natural language and have shown promise in extending to other symbolic domains like coding or mathematics. However, the extent to which they implicitly model symbolic music remains underexplored. This paper investigates how LLMs represent musical concepts by generating symbolic music data from textual prompts describing combinations of genres and styles, and evaluating their utility through recognition and generation tasks. We produce a dataset of LLM-generated MIDI files without relying on explicit musical training. We then train neural networks entirely on this LLM-generated MIDI dataset and perform genre and style classification as well as melody completion, benchmarking their performance against established models. Our results demonstrate that LLMs can infer rudimentary musical structures and temporal relationships from text, highlighting both their potential to implicitly encode musical patterns and their limitations due to a lack of explicit musical context, shedding light on their generative capabilities for symbolic music.

Andrew Shin, Masato Ishii, Takuya Narihira

Transformer architectures have brought about fundamental changes to computational linguistic field, which had been dominated by recurrent neural networks for many years. Its success also implies drastic changes in cross-modal tasks with language and vision, and many researchers have already tackled the issue. In this paper, we review some of the most critical milestones in the field, as well as overall trends on how transformer architecture has been incorporated into visuolinguistic cross-modal tasks. Furthermore, we discuss its current limitations and speculate upon some of the prospects that we find imminent.

Takuya Narihira, Javier Alonsogarcia, Fabien Cardinaux et al.

While there exist a plethora of deep learning tools and frameworks, the fast-growing complexity of the field brings new demands and challenges, such as more flexible network design, speedy computation on distributed setting, and compatibility between different tools. In this paper, we introduce Neural Network Libraries (https://nnabla.org), a deep learning framework designed from engineer's perspective, with emphasis on usability and compatibility as its core design principles. We elaborate on each of our design principles and its merits, and validate our attempts via experiments.

Naofumi Akimoto, Akio Hayakawa, Andrew Shin et al.

We propose a novel reference-based video colorization framework with spatiotemporal correspondence. Reference-based methods colorize grayscale frames referencing a user input color frame. Existing methods suffer from the color leakage between objects and the emergence of average colors, derived from non-local semantic correspondence in space. To address this issue, we warp colors only from the regions on the reference frame restricted by correspondence in time. We propagate masks as temporal correspondences, using two complementary tracking approaches: off-the-shelf instance tracking for high performance segmentation, and newly proposed dense tracking to track various types of objects. By restricting temporally-related regions for referencing colors, our approach propagates faithful colors throughout the video. Experiments demonstrate that our method outperforms state-of-the-art methods quantitatively and qualitatively.

Andrew Shin, Yoshitaka Ushiku, Tatsuya Harada

Image description task has been invariably examined in a static manner with qualitative presumptions held to be universally applicable, regardless of the scope or target of the description. In practice, however, different viewers may pay attention to different aspects of the image, and yield different descriptions or interpretations under various contexts. Such diversity in perspectives is difficult to derive with conventional image description techniques. In this paper, we propose a customized image narrative generation task, in which the users are interactively engaged in the generation process by providing answers to the questions. We further attempt to learn the user's interest via repeating such interactive stages, and to automatically reflect the interest in descriptions for new images. Experimental results demonstrate that our model can generate a variety of descriptions from single image that cover a wider range of topics than conventional models, while being customizable to the target user of interaction.

Andrew Shin, Leopold Crestel, Hiroharu Kato et al.

Automatic melody generation for pop music has been a long-time aspiration for both AI researchers and musicians. However, learning to generate euphonious melody has turned out to be highly challenging due to a number of factors. Representation of multivariate property of notes has been one of the primary challenges. It is also difficult to remain in the permissible spectrum of musical variety, outside of which would be perceived as a plain random play without auditory pleasantness. Observing the conventional structure of pop music poses further challenges. In this paper, we propose to represent each note and its properties as a unique `word,' thus lessening the prospect of misalignments between the properties, as well as reducing the complexity of learning. We also enforce regularization policies on the range of notes, thus encouraging the generated melody to stay close to what humans would find easy to follow. Furthermore, we generate melody conditioned on song part information, thus replicating the overall structure of a full song. Experimental results demonstrate that our model can generate auditorily pleasant songs that are more indistinguishable from human-written ones than previous models.

Andrew Shin, Yoshitaka Ushiku, Tatsuya Harada

Visual Question Answering (VQA) task has showcased a new stage of interaction between language and vision, two of the most pivotal components of artificial intelligence. However, it has mostly focused on generating short and repetitive answers, mostly single words, which fall short of rich linguistic capabilities of humans. We introduce Full-Sentence Visual Question Answering (FSVQA) dataset, consisting of nearly 1 million pairs of questions and full-sentence answers for images, built by applying a number of rule-based natural language processing techniques to original VQA dataset and captions in the MS COCO dataset. This poses many additional complexities to conventional VQA task, and we provide a baseline for approaching and evaluating the task, on top of which we invite the research community to build further improvements.

Kuniaki Saito, Andrew Shin, Yoshitaka Ushiku et al.

Visual question answering (VQA) task not only bridges the gap between images and language, but also requires that specific contents within the image are understood as indicated by linguistic context of the question, in order to generate the accurate answers. Thus, it is critical to build an efficient embedding of images and texts. We implement DualNet, which fully takes advantage of discriminative power of both image and textual features by separately performing two operations. Building an ensemble of DualNet further boosts the performance. Contrary to common belief, our method proved effective in both real images and abstract scenes, in spite of significantly different properties of respective domain. Our method was able to outperform previous state-of-the-art methods in real images category even without explicitly employing attention mechanism, and also outperformed our own state-of-the-art method in abstract scenes category, which recently won the first place in VQA Challenge 2016.

Andrew Shin, Katsunori Ohnishi, Tatsuya Harada

Recent advances in image captioning task have led to increasing interests in video captioning task. However, most works on video captioning are focused on generating single input of aggregated features, which hardly deviates from image captioning process and does not fully take advantage of dynamic contents present in videos. We attempt to generate video captions that convey richer contents by temporally segmenting the video with action localization, generating multiple captions from multiple frames, and connecting them with natural language processing techniques, in order to generate a story-like caption. We show that our proposed method can generate captions that are richer in contents and can compete with state-of-the-art method without explicitly using video-level features as input.

Andrew Shin, Masataka Yamaguchi, Katsunori Ohnishi et al.

The workflow of extracting features from images using convolutional neural networks (CNN) and generating captions with recurrent neural networks (RNN) has become a de-facto standard for image captioning task. However, since CNN features are originally designed for classification task, it is mostly concerned with the main conspicuous element of the image, and often fails to correctly convey information on local, secondary elements. We propose to incorporate coding with vector of locally aggregated descriptors (VLAD) on spatial pyramid for CNN features of sub-regions in order to generate image representations that better reflect the local information of the images. Our results show that our method of compact VLAD coding can match CNN features with as little as 3% of dimensionality and, when combined with spatial pyramid, it results in image captions that more accurately take local elements into account.