Animal sound classification (ASC) refers to the automatic identification of animal categories by sound, and is useful for monitoring rare or elusive wildlife. Thus far, deep-learning-based models have shown good performance in ASC when training data is sufficient, but suffer from severe performance degradation if not. Recently, generative adversarial networks (GANs) have shown the potential to solve this problem by generating virtual data. However, in a multi-class environment, existing GAN-based methods need to construct separate generative models for each class. Additionally, they only consider the waveform or spectrogram of sound, resulting in poor quality of the generated sound. To overcome these shortcomings, we propose a two-step sound augmentation scheme using a class-conditional GAN. First, common features are learned from all classes of animal sounds, and multiple classes of animal sounds are generated based on the features that consider both waveforms and spectrograms using class-conditional GAN. Second, we select data from the generated data based on the confidence of the pretrained ASC model to improve classification performance. Through experiments, we show that the proposed method improves the accuracy of the basic ASC model by up to 18.3%, which corresponds to a performance improvement of 13.4% compared to the second-best augmentation method.

In order to minimize the impact of influenza on public health, accurate early forecasting is essential. Various deep-learning-based models have been proposed to predict future influenza occurrences by capturing temporal/regional patterns from past occurrence time-series data. However, the prediction performance of these unimodal approaches is limited because they extract knowledge only from collected data, and users cannot input contextual information and domain knowledge to them. Recently, large language models (LLMs) have demonstrated the potential to improve prediction accuracy by linking contextual text information to time-series predictions. In this paper, we propose MIFlu, a multimodal influenza forecasting scheme that can fuse contextual text information to time-series influenza occurrences using two LLMs. It first extracts text embeddings from the user's text prompts that contain contextual information using a text-embedding LLM. Then, MIFlu fuses the text embeddings and time-series embeddings and uses the fused embeddings to predict future occurrences using a forecasting LLM. In extensive experiments using public national/regional influenza datasets, MIFlu outperforms other predictive models, improving prediction performance by up to 26.2% compared to state-of-the-art models. We also analyze the effect of various textual input embedders, hyperparameters, and the amount of training data on forecasting accuracy.

Deep learning models that require vast amounts of training data struggle to achieve good animal sound classification (ASC) performance. Among recent few-shot ASC methods to address the data shortage problem regarding animals that are difficult to observe, model-agnostic meta-learning (MAML) has shown new possibilities by encoding common prior knowledge derived from different tasks into the model parameter initialization of target tasks. However, when the knowledge on animal sounds is difficult to generalize due to its diversity, MAML exhibits poor ASC performance due to the static initialization setting. In this paper, we propose a novel task-adaptive parameter transformation scheme called few-shot ASC. TAPT generates transformation variables while learning common knowledge and uses the variables to make parameters specific to the target task. Owing to this transformation, TAPT can reduce overfitting and enhance adaptability, training speed, and performance in heterogeneous tasks compared to MAML. In experiments on two public datasets on the same backbone network, we show that TAPT outperforms the existing few-shot ASC schemes in terms of classification accuracy, and in particular a performance improvement of 20.32% compared to the state-of-the-art scheme. In addition, we show that TAPT is robust to hyperparameters and efficient for training.

Conditional floor plan generation must satisfy constraints comprising user-specified input conditions and design rules. Existing large language model (LLM)-based methods encode floor plans with general text tokens not optimized for this task or apply reinforcement learning (RL) with only one or two coarse sequence-level rewards, failing to generate constraint-faithful floor plans. This work proposes FloorplanLLM, a framework based on domain-specific tokens and RL with verifiable rewards. Floor plan-specific custom tokens are incorporated into the pretrained LLM vocabulary, and floor plans are represented as sequences of these tokens to reduce reliance on general text tokens. Verifiable reward functions evaluate constraint satisfaction in generated floor plans, aligning the LLM to diverse constraints during RL. Token-level credit assignment differentiates reward signals across tokens according to their contribution. Comparative experiments demonstrate approximately 70% higher floor plan quality and more realistic, constraint-faithful plans than existing methods. Ablation studies confirm the contributions of each component.

Conditional floor plan generation must satisfy constraints comprising user-specified input conditions and design rules. Existing large language model (LLM)-based methods use general text tokens that encode floor plans inefficiently or apply reinforcement learning (RL) with only one or two coarse sequence-level rewards, limiting enforcement of diverse constraints on generated plans. This work proposes FloorplanLLM, a framework based on domain-specific tokens and RL with verifiable rewards. Floor plan-specific custom tokens are incorporated into the pretrained LLM vocabulary, and floor plans are represented as sequences of these tokens to reduce reliance on general text tokens. Verifiable reward functions evaluate constraint satisfaction in generated floor plans, aligning the LLM to diverse constraints during RL. Token-level credit assignment differentiates reward signals across tokens according to their contribution. Comparative experiments demonstrate up to 61.9% lower Fréchet Inception Distance score and more realistic, constraint-satisfying plans than existing methods. Ablation studies confirm the contributions of each component.