Selected work

Representative Papers

Analyzing RV32/RV64 Trade-offs for FreeRTOS Latency on 8-Stage RISC-V Soft Processors

Jul 31, 2026

This study addresses the lack of systematic guidance for implementing preemptive real-time operating systems (RTOS) on custom bare-metal RISC-V soft cores, particularly regarding the interplay among processor microarchitecture, interrupt handling, and context switching. The authors deploy FreeRTOS on an 8-stage pipelined RV32/RV64 soft core featuring machine-mode CSRs, a CLINT timer, and exception control logic, and—within a unified microarchitectural and firmware framework—employ Rhealstone-derived microbenchmarks to comparatively evaluate task-switching and preemption latencies between RV32 and RV64. Results reveal that RV64 incurs 36.6% and 17.7% higher cycle delays due to its doubled trap frame size, 64-bit pointers, and scheduler overhead. The complete RTL, firmware, and benchmark suite are open-sourced, establishing a reproducible baseline for RISC-V real-time system design.

0 citationsRead paper

When Should LLMs Search? Counterfactual Supervision for Search Routing

Jul 06, 2026

Large language models often suffer from suboptimal task performance due to inaccurate decisions about when to invoke external search. This work formulates search invocation as an instance-level routing problem and introduces three counterfactual oracles—no-search, search, and unsolvable—defined by task-specific success criteria to generate supervised signals for unified evaluation and training. Leveraging supervised fine-tuning and preference optimization guided by these success metrics, the proposed approach substantially improves routing accuracy. Experiments demonstrate that macro F1 scores for routing increase from 0.7082 to 0.8235 on Gemma E2B and from 0.7053 to 0.8365 on Qwen3.5-4B, effectively reducing both unnecessary and omitted search calls.

0 citationsRead paper

RV-IM100: Quantifying ISA Extension, Datapath Width, and Pipeline Depth Trade-offs in RISC-V Microarchitectures

May 03, 2026

This study addresses the lack of systematic empirical analysis on the multidimensional design trade-offs among instruction set extensions, datapath width, and pipeline depth in RISC-V microarchitectures under a unified baseline. Building upon a common five-stage pipeline, the authors implement ten incrementally evolved FPGA prototypes, co-varying RV32/RV64 datapaths, I/IM instruction sets, and pipeline depths from five to eight stages. An iterative timing-convergence methodology combined with Dhrystone and CoreMark benchmarks enables controlled evaluation. The work reveals, for the first time under such conditions, that frequency gains do not translate linearly into throughput improvements, that RV32 consistently outperforms RV64 in both absolute performance and resource efficiency, and that an eight-stage pipeline markedly reduces resource utilization. Notably, RV32IM achieves a frequency increase from 43 MHz to 126 MHz, yielding 64% higher Dhrystone throughput and a 300% improvement in CoreMark. All RTL and configuration files are publicly released.

0 citationsRead paper

The Effect of Empathic Expression Levels in Virtual Human Interaction: A Controlled Experiment

Dec 23, 2025

This study investigates how varying intensities of empathic expression in virtual humans affect user experience in emotional counseling scenarios. Moving beyond binary empathic presence/absence, empathy is modeled as a tunable continuous variable. Three experimental conditions are designed: neutral dialogue, conversational empathy, and video-driven empathy (generated via the Facial Action Coding System). Subjective self-report scales and one-way ANOVA are employed for evaluation. Results indicate that video-driven empathy significantly enhances users’ affective empathy (*p* < .001), facial naturalness, and expressive appropriateness; however, no significant improvement is observed in cognitive empathy. The core contribution is the first empirical validation of the decisive role of visual embodiment cues—particularly dynamic facial expressions—in eliciting affective empathy. Moreover, the study establishes empathy intensity as a viable and effective design dimension for empathic virtual agents, thereby advancing both theoretical modeling and practical interface design in affective human–computer interaction.

0 citationsRead paper

Design Framework for Conversational Agent in Couple relationships: A Systematic Review

Oct 19, 2025

Current conversational agent research predominantly focuses on individual psychological interventions, leaving a critical gap in the design of emotion-support systems for romantic partnerships. Method: We conducted a PRISMA-compliant systematic review (12 empirical studies) and thematic analysis, integrating attachment theory, relational systems theory, and human-AI interaction design principles. Contribution/Results: We propose the first relationship-centered conversational agent design paradigm, introducing the novel “relationship expert” role. This framework articulates eight evidence-informed design guidelines spanning multimodal interaction, ethical safety, emotional competence, multi-stakeholder alliance building, cultural responsiveness, and relational emotional safety. It bridges theoretical and practical gaps in couple-oriented AI mental health interventions and provides an actionable, theoretically grounded blueprint for developing next-generation agents capable of relational awareness and dynamic adaptation.

0 citationsRead paper

Recent publications

Latest Papers

Analyzing RV32/RV64 Trade-offs for FreeRTOS Latency on 8-Stage RISC-V Soft Processors

Jul 31, 2026

This study addresses the lack of systematic guidance for implementing preemptive real-time operating systems (RTOS) on custom bare-metal RISC-V soft cores, particularly regarding the interplay among processor microarchitecture, interrupt handling, and context switching. The authors deploy FreeRTOS on an 8-stage pipelined RV32/RV64 soft core featuring machine-mode CSRs, a CLINT timer, and exception control logic, and—within a unified microarchitectural and firmware framework—employ Rhealstone-derived microbenchmarks to comparatively evaluate task-switching and preemption latencies between RV32 and RV64. Results reveal that RV64 incurs 36.6% and 17.7% higher cycle delays due to its doubled trap frame size, 64-bit pointers, and scheduler overhead. The complete RTL, firmware, and benchmark suite are open-sourced, establishing a reproducible baseline for RISC-V real-time system design.

0 citationsRead paper

When Should LLMs Search? Counterfactual Supervision for Search Routing

Jul 06, 2026

Large language models often suffer from suboptimal task performance due to inaccurate decisions about when to invoke external search. This work formulates search invocation as an instance-level routing problem and introduces three counterfactual oracles—no-search, search, and unsolvable—defined by task-specific success criteria to generate supervised signals for unified evaluation and training. Leveraging supervised fine-tuning and preference optimization guided by these success metrics, the proposed approach substantially improves routing accuracy. Experiments demonstrate that macro F1 scores for routing increase from 0.7082 to 0.8235 on Gemma E2B and from 0.7053 to 0.8365 on Qwen3.5-4B, effectively reducing both unnecessary and omitted search calls.

0 citationsRead paper

RV-IM100: Quantifying ISA Extension, Datapath Width, and Pipeline Depth Trade-offs in RISC-V Microarchitectures

May 03, 2026

This study addresses the lack of systematic empirical analysis on the multidimensional design trade-offs among instruction set extensions, datapath width, and pipeline depth in RISC-V microarchitectures under a unified baseline. Building upon a common five-stage pipeline, the authors implement ten incrementally evolved FPGA prototypes, co-varying RV32/RV64 datapaths, I/IM instruction sets, and pipeline depths from five to eight stages. An iterative timing-convergence methodology combined with Dhrystone and CoreMark benchmarks enables controlled evaluation. The work reveals, for the first time under such conditions, that frequency gains do not translate linearly into throughput improvements, that RV32 consistently outperforms RV64 in both absolute performance and resource efficiency, and that an eight-stage pipeline markedly reduces resource utilization. Notably, RV32IM achieves a frequency increase from 43 MHz to 126 MHz, yielding 64% higher Dhrystone throughput and a 300% improvement in CoreMark. All RTL and configuration files are publicly released.

0 citationsRead paper

The Effect of Empathic Expression Levels in Virtual Human Interaction: A Controlled Experiment

Dec 23, 2025

This study investigates how varying intensities of empathic expression in virtual humans affect user experience in emotional counseling scenarios. Moving beyond binary empathic presence/absence, empathy is modeled as a tunable continuous variable. Three experimental conditions are designed: neutral dialogue, conversational empathy, and video-driven empathy (generated via the Facial Action Coding System). Subjective self-report scales and one-way ANOVA are employed for evaluation. Results indicate that video-driven empathy significantly enhances users’ affective empathy (*p* < .001), facial naturalness, and expressive appropriateness; however, no significant improvement is observed in cognitive empathy. The core contribution is the first empirical validation of the decisive role of visual embodiment cues—particularly dynamic facial expressions—in eliciting affective empathy. Moreover, the study establishes empathy intensity as a viable and effective design dimension for empathic virtual agents, thereby advancing both theoretical modeling and practical interface design in affective human–computer interaction.

0 citationsRead paper

Design Framework for Conversational Agent in Couple relationships: A Systematic Review

Oct 19, 2025

Current conversational agent research predominantly focuses on individual psychological interventions, leaving a critical gap in the design of emotion-support systems for romantic partnerships. Method: We conducted a PRISMA-compliant systematic review (12 empirical studies) and thematic analysis, integrating attachment theory, relational systems theory, and human-AI interaction design principles. Contribution/Results: We propose the first relationship-centered conversational agent design paradigm, introducing the novel “relationship expert” role. This framework articulates eight evidence-informed design guidelines spanning multimodal interaction, ethical safety, emotional competence, multi-stakeholder alliance building, cultural responsiveness, and relational emotional safety. It bridges theoretical and practical gaps in couple-oriented AI mental health interventions and provides an actionable, theoretically grounded blueprint for developing next-generation agents capable of relational awareness and dynamic adaptation.

0 citationsRead paper