arXiv subjects

Enbo Zhang

Publications and source records attributed to Enbo Zhang.

4 recordsLinked to original sources

arXiv2026

Reasoning with Memory: A Temporal Granularity-Adaptive Framework for Training-Free Long Video Understanding

While Multimodal Large Language Models (MLLMs) demonstrate superior generalization in fundamental video tasks, restricted context windows limit their long video understanding. To accommodate this constraint, models typically resort to keyframe selection. However, uniform sampling or static query-guided selection often overlooks critical temporal context, failing to adapt to the varying query temporal granularities. In this paper, we propose ReMem, a temporal granularity-adaptive keyframe selection framework for training-free LongVideoQA. ReMem introduces a dual-level memory-augmented adaptation. At the query level, Memory-Driven Question Parsing leverages LLM long-term memory to decode question temporal granularity and extract semantic entities. At the video level, Synergistic Dual-Semantic Frame Alignment exploits intrinsic structural memory to align frames with query semantics, guiding Structure-Aware Dynamic Frame Routing to cluster events and optimally distribute sampling budgets. By explicitly preserving temporal information with memory mechanisms, ReMem suppresses redundancy and empowers MLLMs to perform robust multi-granular video reasoning. Evaluations across four popular LongVideoQA benchmarks using three MLLMs demonstrate highly efficient, state-of-the-art zero-shot performance; notably, LLaVA-Video with ReMem reaches 54.5% (+12.3%) on LVBench and 67.1% (+8.2%) on LongVideoBench.

cs.AI↗

arXiv2026

Multi-task deep neural network for predicting both nuclear fission yields and their experimental errors in peak-shaped data

The fission product yield (FPY) is crucially important information for numerous nuclear applications. However, the peak-shaped characteristics of FPY data present important challenges for predicting unobservable FPY data. To address these challenges, after applying Multi-task learning models to fission product yield data and their experimental error estimates, we introduce a novel loss function along with incorporation of the odd even effect. Our approach is intended to predict unknown fission yields and the associated experimental error. To demonstrate the effectiveness of our proposed method, we compared our proposed method with conventional methods that learn each dataset independently. Our findings demonstrate that the proposed methods can predict peak shaped data with experimental error estimates more effectively than earlier methods can.

nucl-th↗

arXiv2024

Ultralow thermal conductance across the [FePt/h-BN/FePt] interface

Heat transfer in nanocomposite materials has attracted great interest for various applications. Multilayer structures provide an important platform to study interfacial thermal transport and to engineer materials with ultralow thermal conductivity. Here we report on the fabrication and thermal characterization of [h-BN/$L1_0$-FePt]xN multilayers, where hexagonal boron nitride (h-BN) nanosheets (2.5 nm thick) and $L1_0$-FePt layers (6.5 nm thick) alternate periodically. Differential three-omega($3ω$) measurements reveal an ultralow effective thermal conductivity of $ 0.60 \pm 0.05 W \cdot m^{-1}K^{-1}$ across the multilayer films, and a low thermal boundary conductance (TBC) of $ 67.9 \pm 6.6 MW \cdot m^{-2}K^{-1}$ for the [FePt/h-BN(2.5nm)/FePt] interface at room temperature. We attribute the ultralow thermal conductivity to the weak van der Waals bonding at h-BN/FePt interfaces, which dominates the thermal resistance of the multilayer structure. These findings provide insights into the thermal transport in 2D-material/metal multilayer nanostructures and suggest the [h-BN/FePt] superlattice as a promising material for nanoscale thermal barrier coating. Furthermore, the obtained TBC lays the foundation for analyzing heat transfer in FePt-(h-BN) nanogranular films, a promising magnetic recording media which can potentially provide high thermal gradient for heat-assisted magnetic recording (HAMR). This work advances the understanding of thermal transport in 2D-material/metal nanocomposites and demonstrates interface engineering as an effective approach to achieve materials with ultralow thermal conductivity.

physics.app-ph↗

arXiv2019

Iridium Enabled Field-free Spin-orbit Torque Switching of Perpendicular Magnetic Tunnel Junction Device

Writing magnetic bits by spin-orbit torques (SOTs) arising from spin Hall effect creates new possibilities for ultrafast and low-power magnetoresistive random access memory (MRAM). For perpendicular MRAM, an extra in-plane field is required to break the symmetry for the deterministic SOT writing of the perpendicular storage layer. Although schemes have been demonstrated in external-field-free SOT switching of a perpendicular layer, practically integrating them with perpendicular MTJs still appears to be challenging. Here, we present experimental demonstration of spin-orbit torques (SOTs) switching a perpendicular magnetic tunnel junction (MTJ) device without applying an external magnetic field. An Ir layer is used to serve dual-purpose of both injecting the pure spin current via spin Hall effect and mediating an in-plane exchange field to the perpendicular free layer of the MTJ. Robust field-free SOT switching with pulsed write path current is demonstrated for various MTJ sizes ranging from 50 nm to 500 nm. The effect of MTJ size and pulse width on the critical switching current is studied. Combined micromagnetic simulations are carried out to provide in-depth analysis of the switching dynamics as well as the thermal effect on the switching.

physics.app-ph↗