Reference-based annotation tools have become standard for cell type assignment in single-cell RNA sequencing, leveraging large-scale atlases to transfer labels to new datasets. However, a substantial fraction of cells often receive uncertain or ambiguous annotations-low confidence scores, competing label probabilities, or high entropy across cell types. These cells are typically treated as technical artifacts and filtered out, yet in perturbation experiments they may represent the biologically interesting deviations that investigators seek to identify. We present AnnQ (Annotation Quantification of cellular identity uncertainty), a Python framework that repurposes annotation uncertainty as a quantitative measure of cellular abnormality. AnnQ extracts uncertainty-aware features from probabilistic cell type assignments-including confidence, confidence gap, admixture ratio, and entropy-and computes an out-of-reference (OOR) score measuring each cell's deviation from a reference population in multivariate uncertainty space. Applying AnnQ to genetic perturbation and drug resistance datasets, we show that OOR scores detect aberrant cellular states that are not resolved by conventional clustering or differential abundance analyses. AnnQ provides a complementary approach for characterizing transitional and abnormal cell states at single-cell resolution. AnnQ is implemented in Python, and its source code and documentation are available on https://github.com/joonan-lab/AnnQ.git.

SUMMARY: Reconstructing gene regulatory networks (GRNs) from single-cell RNA sequencing (scRNA-seq) data is fundamental for understanding cellular dynamics at the molecular level but requires sophisticated workflows. Here, we introduce CellCraft, a web-based application designed to streamline GRN inference. CellCraft integrates multiple GRN reconstruction tools, including TENET, within a unified web application featuring an intuitive graphical user interface. Notably, CellCraft provides a visual programming interface that simplifies the design and execution of complex multistep analyses, thereby enhancing accessibility and facilitating the visualization and interpretation of computational experiments. Furthermore, its modular plugin architecture ensures extensibility, enabling the incorporation of newly developed single-cell analysis algorithms. Consequently, CellCraft provides a user-friendly and extensible application for integrative GRN analysis of scRNA-seq datasets. AVAILABILITY AND IMPLEMENTATION: CellCraft is available on GitHub at https://github.com/cxinsys/cellcraft. The source code has been archived on Zenodo at 10.5281/zenodo.17865848.

Different from color correction and transfer, color grading involves adjusting colors for artistic or storytelling purposes in a video, which is used to establish a specific look or mood. However, due to the complexity of the process and the need for specialized editing skills, video color grading remains primarily the domain of professional colorists. In this paper, we present a reference-based video color grading framework. Our key idea is explicitly generating a look-up table (LUT) for color attribute alignment between reference scenes and input video via a diffusion model. As a training objective, we enforce that high-level features of the reference scenes like look, mood, and emotion should be similar to that of the input video. Our LUT-based approach allows for color grading without any loss of structural details in the whole video frames as well as achieving fast inference. We further build a pipeline to incorporate a user-preference via text prompts for low-level feature enhancement such as contrast and brightness, etc. Experimental results, including extensive user studies, demonstrate the effectiveness of our approach for video color grading. Codes are publicly available at https://github.com/seunghyuns98/VideoColorGrading.

SUMMARY: SCODE reconstructs gene regulatory networks from single-cell RNA sequencing (scRNA-seq) data using an ordinary differential equation (ODE) model, and has been successfully applied to a wide range of scRNA-seq datasets, including mouse, human, and plant cells. However, its computational performance is limited when processing large datasets due to its sequential execution flow and repeated optimization loops. To overcome this limitation, we have developed FastSCODE, a batch computing version of the SCODE algorithm optimized for acceleration on manycore processors such as GPUs. FastSCODE performs batch computation on multiple gene expression profiles and optimizes the parameters of a linear ODE model using manycore computing. Compared to the original implementation, FastSCODE achieves up to 6000× improvement in performance (from about one month to 10 min) on the CeNGEN scRNA-seq dataset when using multiple GPUs. AVAILABILITY AND IMPLEMENTATION: FastSCODE is publicly available on GitHub at https://github.com/cxinsys/fastscode.

NAND flash memory has a disadvantage in that additional work in the flash translation layer (FTL) is required for compatibility with the block interface. In particular, FTL should periodically perform garbage collection (GC) to reclaim free data blocks. Unfortunately, GC includes an expensive erase operation and can cause write amplification (WA), which writes more pages than the system requested. Therefore, it is essential to design a victim selection policy to reduce WA during GC. Greedy and Cost-Benefit (CB) are the most widely known victim selection policies. However, Greedy does not consider locality, and CB suffers computational overhead. This paper proposes Low Computational Cost-Benefit (LC-CB), a novel victim selection policy compensating for these shortcomings. Unlike the existing methods to improve CB, LC-CB changes the operation itself used in the victim selection metric to low computational. This paper describes the constraint required to make a victim selection policy with a low computational overhead and explains that LCCB can consider locality while satisfying these constraints. The experimental results show that our proposed policy can reduce time overhead by 70% compared to CB and reduce WA by up to 30% compared to Greedy.

Finding an optimal learning trajectory is an important question in educational systems. Existing Artificial Intelligence in Education (AiEd) technologies mostly used indirect methods to make the learning process efficient such as recommending contents based on difficulty adjustment, weakness analysis, learning theory, psychometric analysis, or domain specific rules.

We propose SAINT+, a successor of SAINT which is a Transformer based knowledge tracing model that separately processes exercise information and student response information. Following the architecture of SAINT, SAINT+ has an encoder-decoder structure where the encoder applies self-attention layers to a stream of exercise embeddings, and the decoder alternately applies self-attention layers and encoder-decoder attention layers to streams of response embeddings and encoder output. Moreover, SAINT+ incorporates two temporal feature embeddings into the response embeddings: elapsed time, the time taken for a student to answer, and lag time, the time interval between adjacent learning activities. We empirically evaluate the effectiveness of SAINT+ on EdNet, the largest publicly available benchmark dataset in the education domain. Experimental results show that SAINT+ achieves state-of-the-art performance in knowledge tracing with an improvement of 1.25% in area under receiver operating characteristic curve compared to SAINT, the current state-of-the-art model in EdNet dataset.

Real-time applications with high computational demand, e.g., autonomous driving, are emerging and their complex nature conforms to a DAG(directed acyclic graph) structure. We propose a conditionally optimal parallelization for real-time DAG tasks for global EDF, ensuring complete execution of all tasks within the deadline. To achieve this, we formalize a monotonic increasing property of both tolerance and interference to the parallelization option. Using such properties, we develop a unidirectional search algorithm that can assign parallelization options in polynomial time, which we formally prove the optimality. We observe significant improvement of schedulability through simulation experiment, and then in the following implementation experiment, we demonstrate that the algorithm is practically applicable for real-world use-cases.

Student dropout prediction provides an opportunity to improve student engagement, which maximizes the overall effectiveness of learning experiences. However, researches on student dropout were mainly conducted on school dropout or course dropout, and study session dropout in a mobile learning environment has not been considered thoroughly. In this paper, we investigate the study session dropout prediction problem in a mobile learning environment. First, we define the concept of the study session, study session dropout and study session dropout prediction task in a mobile learning environment. Based on the definitions, we propose a novel Transformer based model for predicting study session dropout, DAS: Deep Attentive Study Session Dropout Prediction in Mobile Learning Environment. DAS has an encoder-decoder structure which is composed of stacked multi-head attention and point-wise feed-forward networks. The deep attentive computations in DAS are capable of capturing complex relations among dynamic student interactions. To the best of our knowledge, this is the first attempt to investigate study session dropout in a mobile learning environment. Empirical evaluations on a large-scale dataset show that DAS achieves the best performance with a significant improvement in area under the receiver operating characteristic curve compared to baseline models.

In this paper, we propose a novel Transformer-based model for knowledge tracing, SAINT: Separated Self-AttentIve Neural Knowledge Tracing. SAINT has an encoder-decoder structure where the exercise and response embedding sequences separately enter, respectively, the encoder and the decoder. The encoder applies self-attention layers to the sequence of exercise embeddings, and the decoder alternately applies self-attention layers and encoder-decoder attention layers to the sequence of response embeddings. This separation of input allows us to stack attention layers multiple times, resulting in an improvement in area under receiver operating characteristic curve (AUC). To the best of our knowledge, this is the first work to suggest an encoder-decoder model for knowledge tracing that applies deep self-attentive layers to exercises and responses separately. We empirically evaluate SAINT on a large-scale knowledge tracing dataset, EdNet, collected by an active mobile education application, Santa, which has 627,347 users, 72,907,005 response data points as well as a set of 16,175 exercises gathered since 2016. The results show that SAINT achieves state-of-the-art performance in knowledge tracing with an improvement of 1.8% in AUC compared to the current state-of-the-art model.

Human-machine interaction is one of the most crucial aspects of advanced manufacturing systems that have advanced to so-called smart manufacturing systems. In this regard, this paper presents a framework for formalizing human-machine manufacturing systems. The human-machine system considered in this paper consists of the following three main components: a human supervisor; several cells, each of which is composed of a human operator and a machine; and interfaces. A human operator interacts with a machine in a cell and performs manufacturing tasks based on commands given by the supervisor. Meanwhile, the supervisor is responsible for performing exception handling tasks in response to unanticipated events reported by the cells. With the proposed model, desirable specifications are constructed, which include a condition free of mode confusion, manufacturing task goal reachability, and exception handling task supportability in human- machine manufacturing systems. It is also suggested that adaptive automation with varying levels of information abstraction to humans can be accommodated by the proposed framework. As an illustrative example, we demonstrate the formal models and specifications and the applicability of adaptive automation with a case study of a simple chair assembly system.

Current technology is not sufficient to automate all desired tasks. Human–machine interaction (HMI) has thus become a key control and design factor for tasks requiring human-level decision-making or information synthesis. Such processes require a formal representation of human actions (including decision-making) when modeling HMI systems; however, successful prescriptive approaches to this end have still been elusive. This article extends the affordance-based finite state automata model, conditioning human prior experience and natural memory decay of task knowledge (or skill decay). The new model draws upon both reinforcement learning and natural memory decay for decision-making on action choice. An empirical study is carried out to specify how action choice is affected or updated by reinforcement learning based on past experience, and Wickelgren’s decay function is jointly employed to predict human decision-making behavior.

Recent rich services provided by mobile devices take a form of composite services consisting of several constituent atomic services and they are gaining popularity due to technological advances and wireless network infrastructures. To make the best use of these technological benefits, the services given by the device need to be developed and provided with consideration of user context that contains information about environments. We regard these pieces of information as data segments that can trigger a service when they are gathered enough to infer a certain degree of the user context. In this regard, we investigate a context-aware composite service provision process and present an analytical tool for assessing the performance of the service provision process by means of a mathematical model. Based on the results from extensive experimental simulations, it is observed that the performance assessment based on the proposed mean value analysis (MVA) model effectively confirms the characteristics of the composite service systems.

This paper presents a formal approach to resolve an important question concerning changes in the control of computerized manufacturing systems when a human operator is involved as a task-performing agent. It requires building a model of human functional specifications used in executing tasks and integrating it into a control scheme for the model. More importantly, analysis of control complexity needs to be conducted to build an effective control mechanism. In this paper, a human material handler is considered, and an assessment of part flow complexity affected by human tasks in a highly automated manufacturing system is presented. For this purpose, a formal model of human task-performing processes is proposed in terms of a part and location(s) of a task. A classification for human material handling tasks is presented based on the proposed model. Furthermore, human errors and the impact of human errors on part flow are considered. Part flow complexity of a manufacturing system from the control perspective is then investigated in terms of the human tasks and errors. A shop floor control example where a human operator performs material handling tasks is provided to illustrate the proposed model.

To achieve an effective integration framework for manufacturing systems, a formal model of a system is highly desired. In spite of significant work on automated manufacturing systems, human operators still play a critical role in virtually every system, especially for material-handling processes. To build a model for control and analysis of a system where a human operator is integrated, a formal functional specification of a human material handler (MH) is presented in a hierarchical framework. Two types of human operational errors associated with material-handling tasks are also classified and discussed. A shop floor control example is provided to illustrate the proposed modeling framework.