Understanding human emotions from images is a challenging yet essential task for vision-language models. While recent efforts have fine-tuned vision-language models to enhance emotional awareness, most approaches rely on global visual representations and fail to capture the nuanced and multi-faceted nature of emotional cues. Furthermore, most existing approaches adopt instruction tuning, which requires costly dataset construction and involves training a large number of parameters, thereby limiting their scalability and efficiency. To address these challenges, we propose MASP, a novel framework for Multi-Aspect guided emotion reasoning with Soft Prompt tuning in vision-language models. MASP explicitly separates emotion-relevant visual cues via multi-aspect cross-attention modules and guides the language model using soft prompts, enabling efficient and scalable task adaptation without modifying the base model. Our method achieves state-of-the-art performance on various emotion recognition benchmarks, demonstrating that the explicit modeling of multi-aspect emotional cues with soft prompt tuning leads to more accurate and interpretable emotion reasoning in vision-language models.

We have developed and maintained a set of closely related compilers. Although much of their code is duplicated and shared, they have been maintained separately because they are treated as different compilers. Even if they were merged together, the combined code would become too complicated to serve as the base for another extension. We describe our experience to address this problem by adopting the product line engineering paradigm to build a family of compilers. This paradigm encourages developers to focus on developing a set of compilers rather than on developing one particular compiler. We show engineering activities for a family of compilers from product line analysis through product line architecture design to product line component design. Then, we present how to build particular compilers from core assets resulting from the previous activities and how to take advantage of modern programming language technology to organize this task. Our experience demonstrates that the product line engineering as a developing paradigm can ease the construction of a family of compilers.

We present language mechanisms for polymorphic, extensible records and their exact dual, polymorphic sums with extensible first-class cases. These features make it possible to easily extend existing code with new cases. In fact, such extensions do not require any changes to code that adheres to a particular programming style. Using that style, individual extensions can be written independently and later be composed to form larger components. These language mechanisms provide a solution to the expression problem.We study the proposed mechanisms in the context of an implicitly typed, purely functional language PolyR. We give a type system for the language and provide rules for a 2-phase transformation: first into an explicitly typed λ-calculus with record polymorphism, and finally to efficient index-passing code. The first phase eliminates sums and cases by taking advantage of the duality with records.We implement a version of PolyR extended with imperative features and pattern matching - we call this language MLPolyR. Programs in MLPolyR require no type annotations - the implementation employs a reconstruction algorithm to infer all types. The compiler generates machine code (currently for PowerPC) and optimizes the representation of sums by eliminating closures generated by the dual construction.