Channel identifiability (CID) refers to the ability to distinguish among individual channels in time series (TS) modeling. The absence of CID often results in producing identical outputs for identical inputs, disregarding channel-specific characteristics. In this paper, we highlight the importance of CID and propose Channel Normalization (CN), a simple yet effective normalization strategy that enhances CID by assigning distinct affine transformation parameters to each channel. We further extend CN in two ways: 1) Adaptive CN (ACN) dynamically adjusts parameters based on the input TS, improving adaptability in TS models, and 2) Prototypical CN (PCN) introduces a set of learnable prototypes instead of per-channel parameters, enabling applicability to datasets with unknown or varying number of channels and facilitating use in TS foundation models. We demonstrate the effectiveness of CN and its variants by applying them to various TS models, achieving significant performance gains for both non-CID and CID models. In addition, we analyze the success of our approach from an information theory perspective. Code is available at [https://github.com/seunghan96/CN](https://github.com/seunghan96/CN).

Autonomous mobile robots (AMRs) serve diverse purposes across various settings to assist with material handling operations. These AMRs rely on advanced control methods to manage and execute tasks such as motion planning, environmental perception, localization, navigation, and obstacle avoidance. As warehouse environments become increasingly crowded with people, products, and machines, there is a growing demand for AMR sensing technology, which has recently been equipped for safety and operation efficiency. In this paper, the authors propose to leverage optical flow utilizing AMR vision data to incorporate real-time sensory information into the control strategy framed as Markov Decision Processes (MDP). The proposed model investigates how dynamic and real-time sensing information can be incorporated into MDP models via optical flow, capturing AMR’s surroundings to enhance situational awareness and, consequently, the safety of material-handling robots. Moreover, to maximize the model’s validity and practical application, the authors tested it within a physics-based simulation environment developed in Unreal Engine. Experimental results validate that our optical flow-based results outperform the standard MDP model regarding AMR operation efficiency, proving its enhanced effectiveness and reliability.

Multi-behavior recommender systems, rapidly advancing across various domains, utilize plentiful auxiliary interactions on a variety of user behaviors to enhance recommendations for the target behavior, such as purchases. While previous methods have made strides in leveraging such interactions with advanced machine learning methods, they still face challenges in adequately using multi-faceted relationships among behaviors and handling uncertain auxiliary interactions that could potentially lead to purchases or not. In this paper, we propose MuLe (Multi-Grained Graph Learning), a novel graph-based model designed to address these limitations. We design a multi-grained graph learning strategy to capture diverse aspects of behaviors, ranging from unified to specific, and then to target-related behavior interactions. To handle uncertain interactions, we use graph attention, weighting the importance of those interactions related to the target behavior. Afterward, we use an attention mechanism to effectively aggregate diverse behavior embeddings obtained from the multi-grained graph encoders. Extensive experiments show that MuLe significantly outperforms the state-of-the-art methods, achieving improvements of up to 44.6% in HR@10 and 52.9% in NDCG@10, respectively. Our code and datasets are available at https://github.com/geonwooko/MULE.

Contrastive learning has shown to be effective to learn representations from time series in a self-supervised way. However, contrasting similar time series instances or values from adjacent timestamps within a time series leads to ignore their inherent correlations, which results in deteriorating the quality of learned representations. To address this issue, we propose \textit{SoftCLT}, a simple yet effective soft contrastive learning strategy for time series. This is achieved by introducing instance-wise and temporal contrastive loss with soft assignments ranging from zero to one. Specifically, we define soft assignments for 1) instance-wise contrastive loss by distance between time series on the data space, warping and 2) temporal contrastive loss by the difference of timestamps. SoftCLT is a plug-and-play method for time series contrastive learning that improves the quality of learned representations without bells and whistles. In experiments, we demonstrate that SoftCLT consistently improves the performance in various downstream tasks including classification, semi-supervised learning, transfer learning, and anomaly detection, showing state-of-the-art performance. Code is available at this repository: https://github.com/seunghan96/softclt.

Masked time series modeling has recently gained much attention as a self-supervised representation learning strategy for time series. Inspired by masked image modeling in computer vision, recent works first patchify and partially mask out time series, and then train Transformers to capture the dependencies between patches by predicting masked patches from unmasked patches. However, we argue that capturing such patch dependencies might not be an optimal strategy for time series representation learning; rather, learning to embed patches independently results in better time series representations. Specifically, we propose to use 1) the simple patch reconstruction task, which autoencode each patch without looking at other patches, and 2) the simple patch-wise MLP that embeds each patch independently. In addition, we introduce complementary contrastive learning to hierarchically capture adjacent time series information efficiently. Our proposed method improves time series forecasting and classification performance compared to state-of-the-art Transformer-based models, while it is more efficient in terms of the number of parameters and training time. Code is available at this repository: https://github.com/seunghan96/pits.

Advances in diffusion models for generative artificial intelligence have recently propagated to the time series (TS) domain, demonstrating state-of-the-art performance on various tasks. However, prior works on TS diffusion models often borrow the framework of existing works proposed in other domains without considering the characteristics of TS data, leading to suboptimal performance. In this work, we propose Adaptive Noise schedule for Time series diffusion models (ANT), which automatically predetermines proper noise schedules for given TS datasets based on their statistics representing non-stationarity. Our intuition is that an optimal noise schedule should satisfy the following desiderata: 1) It linearly reduces the non-stationarity of TS data so that all diffusion steps are equally meaningful, 2) the data is corrupted to the random noise at the final step, and 3) the number of steps is sufficiently large. The proposed method is practical for use in that it eliminates the necessity of finding the optimal noise schedule with a small additional cost to compute the statistics for given datasets, which can be done offline before training. We validate the effectiveness of our method across various tasks, including TS forecasting, refinement, and generation, on datasets from diverse domains. Code is available at this repository: https://github.com/seunghan96/ANT.

Nonvolatile magnetic RAM (MRAM) offers high cell density and low leakage power. This paper reports on using a 3-D integration technology based on through-silicon vias, to stack disparate memory technologies (e.g., SRAM and MRAM) together onto chip multiprocessors. In this paper, we explore the design of a 3-D stacked nonuniform hybrid SRAM/MRAM L2 cache architecture (NUCA) using the on-chip network to mitigate the interconnection problem. In addition, this paper investigates the problem of partitioning shared SRAM/MRAM hybrid L2 cache and placing cache data into the partitioned 3-D stacked hybrid NUCA for concurrently executing multiple applications in order to improve the system performance in terms of instructions per second while considering the heterogeneous characteristics in interconnection wire delay, memory cell density, memory access latency, and memory power consumption in a 3-D stacked hybrid SRAM/MRAM L2 cache. Experimental results show that the proposed runtime method with a 3-D stacked hybrid L2 cache improves performance by 61%, energy efficiency, i.e., energy-delay product by 53%, and storage lifetime by 15.6 times on average compared with the conventional SRAM-only L2 cache or MRAM-only L2 cache with the similar area.

Nonvolatile memory such as magnetic RAM (MRAM) offers high cell density and low leakage power while suffering from long write latency and high write energy, compared with SRAM. 3-D integration technology using through-silicon vias enables stacking disparate memory technologies (e.g., SRAM and MRAM) together onto chip-multiprocessors (CMPs). The use of hybrid memories as an on-chip cache can take advantage of the best characteristics that each technology offers. However, the inherent high power density and heat removal limitation in 3-D integrated circuits may incur temperature-related problems. In this paper, we propose a runtime thermal management method for CMPs with the 3-D stacked hybrid SRAM/MRAM L2 cache. The proposed method combines dynamic cache management such as resource allocation, way-based power gating, and data migration with dynamic voltage and frequency scaling of processing cores in a temperature- and energy-aware manner. Experimental results show that the proposed runtime method with the 3-D stacked hybrid L2 cache offers up to 107.37% (55.28% on average) performance improvement and 88.47% (47.65% on average) energy efficiency improvement compared with existing thermal management methods with 3-D stacked SRAM-based L2 cache.

Energy consumption is a crucial issue of mobile surveillance cameras owing to limited battery capacity. The lifetime of the system is significantly extended by the event-driven operation; the system mostly stays in sleep mode and wakes up only when an event is detected. In this paper, we propose a design of a low-energy surveillance camera that records events such as the abnormal movement of objects, or physical damage to the camera itself. Unlike conventional event-driven approaches, the proposed system records video from 10 seconds before the event detection because the most critical information is often before or at the moment of event detection, not after the detection. Two different encoders, a low-power encoder and a high-compression encoder, are employed together to implement the low-energy surveillance camera. Experimental results show that the energy consumption of the whole system is reduced by up to 74.9% (by 66.8% on average) compared with conventional always-on system.

As technology scales, increasing capacity of cache memory leads to increase in leakage power dissipation, especially in three-dimensional (3D) IC with high thermal density. In this paper, we explore how cache data can be mapped on a multi-processor architecture in 3D IC to minimize energy consumption with considering temperature distribution and bus traffic congestion. Simulation results based on ILP (Integer Linear Programming) formulation show that the proposed cache data mapping approach achieves up to 30.7% energy reduction compared to the case of considering temperature distribution only.