Combining Mixture of Experts (MoE) with Low-Rank Adaptation (LoRA) has shown promising efficiency in multi-task instruction tuning for Large Language Models (LLMs). While existing routing schemes for such MoE systems employ auxiliary functions to ensure both expert selection certainty and workload balance among experts, they are hindered by two critical challenges: (1) Existing methods overlook the evolving cross-expert relationships across layers, leading to inefficient expert utilization. (2) The auxiliary functions fail to incorporate cross-task semantic characteristics during expert assignment, leading to suboptimal task adaptation. To address these challenges, we propose Hybrid routing for a Mixture of LoRA Experts (HotMoE), a novel multi-task instruction tuning framework that adapts hierarchical routing to the distinct characteristics of different LLM layers. First, we design a hybrid routing module. In lower layers, expert-expert attention facilitates cross-task collaboration and generalization. In higher layers, token-expert attention enables precise alignment between task semantics and specialized experts. Second, we introduce a similarity-guided auxiliary loss module to regularize routing decisions by exploiting hidden state similarities. This loss synergistically reinforces expert specialization without sacrificing certainty of expert selection by promoting cohesive activation patterns among semantically related tasks while sharpening distinctions between conflicting ones. Experiments across two multi-task instruction tuning scenarios covering seven NLP benchmarks demonstrate that HotMoE consistently outperforms all baselines, improving Mean Relative Difference by up to 1.68% with only 3.1% of trainable parameters.

The reputation of universities has drawn increasing attention in recent years, especially with the emergence of various rankings. However, despite advances in big data technologies that facilitate data collection and analysis, accurately defining and balancing factors related to university reputation and educational quality remains complex and tedious. Moreover, current educational assessment methods exhibit notable differences and controversies. In this paper, we present Iva , a human-in-the-loop I ntelligent V isual A ssessment system for higher education quality. This system utilizes large language models to analyze extensive multi-modal educational data, with visualization techniques incorporated to enable multi-scale exploration and interaction. Our extensive evaluations, including a carefully-designed user study and expert interviews, demonstrate the system’s potential value and provide insights for future improvements.

HCI has recently increased its interest in the domains of museums and gifting. The former is often oriented primarily towards the past, while the latter is often oriented towards the future, in terms of anticipating the receiver’s reactions. Our article provides a sustained and well-evidenced new theoretical framework on the role of time-orientation on the design of forward-oriented (gifting) experiences in past-oriented (museum) settings. This Temporal Experience Design Framework develops from the analysis of two such studies, one smartphone app and one VR experience using passive haptics. Both interventions prompted the user to reflect on the past while planning a gift or donation for future consumption. We apply a novel combination of analyses to both projects using the lenses of conversational storytelling, performance, and human geography. Our analyses reveal the power of orienting users towards the past and the future – simultaneously – to enhance the present moment of a performative engagement. Our aim is to provide a conceptual framework that can help design researchers to identify, name, and understand how time-orientation can be used to enhance user and visitor experience. We also extrapolate design guidelines that we expect may be fruitful outside these contexts.

AI and other emerging applications demand domain-specific architectures which require compiler techniques such as back-end generation for different architectures and optimizations. However, traditional undergraduate compiler courses emphasize the front-end, while code generation and optimization are rarely involved. To motivate universities to have more industry-friendly compiler courses, we have designed a national compiler design competition for undergraduates to include compiler techniques beyond parsing. Moreover, we provided reliable, continuous cloud storage and an online evaluation platform for distributed competitors. In the 9-week Competition in 2020, each team (up to 4 students) was required to implement a compiler for a given SysY language and given target hardware (Raspberry Pi 4B). The performance was evaluated by executing code generated by the compiler on real hardware. Finally,21 of 72 teams successfully passed all functional test cases; 12 of 21 teams implemented optimizations showing significant speedup over gcc -O0; furthermore, compilers of the top 3 teams performed better than gcc -O2 on the given 10 performance test cases. Some advanced optimization techniques, such as multithreading and SIMD, were used by some teams. This paper summarizes the competition and further thoughts on compiler courses.

We explore how a combination of manipulations and transitions can extend Substitutional Reality to create a highly personal Virtual Reality experience. Our design aimed to meet two challenges faced by museums: the limitations of object handling and the desire for visitors to create their own interpretations. Using a Research-through-Design methodology, we built a performance-led Mixed Reality (MR) experience that lets museum visitors physically handle 3D prints or scans of museum objects to share personal stories about them. The stories are recorded and donated to the participating museum. We reflect on the complex design and the findings gained from a two-day in-the-wild deployment to explore engagement and disruption through manipulations of physicality, visuals, and scale; the transitions between spaces; and a trajectory of storytelling performance. We chart a wide scope for Performative Substitutional Reality and draw implications for VR, MR, and performance-led research in any context.

Which suggestions for behavioral modifications, based on mathematical models, are most likely to be followed in the real world? We address this question in the context of human circadian rhythms. Jet lag is a consequence of the misalignment of the body's internal circadian (~24-hour) clock during an adjustment to a new schedule. Light is the clock's primary synchronizer. Previous research has used mathematical models to compute light schedules that shift the circadian clock to a new time zone as quickly as possible. How users adjust their behavior when provided with these optimal schedules remains an open question. Here, we report data collected by wearables from more than 100 travelers as they cross time zones using a smartphone app, Entrain. We find that people rarely follow the optimal schedules generated through mathematical modeling entirely, but travelers who better followed the optimal schedules reported more positive moods after their trips. Using the data collected, we improve the optimal schedule predictions to accommodate real-world constraints. We also develop a scheduling algorithm that allows for the computation of approximately optimal schedules "on-the-fly" in response to disruptions. User burnout may not be critically important as long as the first parts of a schedule are followed. These results represent a crucial improvement in making the theoretical results of past work viable for practical use and show how theoretical predictions based on known human physiology can be efficiently used in real-world settings.

Break activities -- deliberate and unexpected -- are common throughout the working day, playing an important role in the wellbeing of workers. This paper investigates the role of increasingly pervasive ICT in creating new opportunities for breaks at work, what impact the technology has on management of boundaries at work, and the effects these changes have on personal wellbeing. We present a study of the routines of office-workers, where we used images from participants' work-days to prompt and contextualize interviews with them. Analysis of coded photographs and interview data makes three contributions: an account of ubiquitous ICT creating new forms of micro-breaks, including the opportunity to employ previously wasted time; a description of the ways in which staff increasingly bring "home to work"; and a discussion of the emergence of "screen guilt". We evaluate our findings in relation to previous studies, and leave three research implications and questions for future work in this domain.