Mask-aware foundational-model embeddings for 18F-FDG-PET/CT prognosis in multiple myeloma.

Cognitive offloading refers to the use of physical actions and the external environment to reduce cognitive demand. Offloading strategies such as creating external reminders instead of relying on internal memory are highly effective and play a key role in supporting real-world cognition. Previous work has shown that people have systematic biases in their offloading strategies, which are related to biased metacognitive evaluations of cognitive ability. While metacognitive interventions could potentially mitigate these biases, research investigating their effects has produced mixed results. Here, we examined the influence of a brief metacognitive intervention comprising just five trials during an initial practice session. After the intervention, participants performed a memory task where they decided between using internal memory (for maximum reward) or external reminders (for reduced reward), allowing us to determine the optimality of offloading strategies. Experiment 1 (N = 164) showed that making metacognitive predictions and subsequently receiving feedback led to improved metacognitive calibration and more optimal reminder-setting strategies. Experiment 2 (N = 416) replicated this pattern and found that making predictions alone was ineffective. These findings suggest that a metacognitive intervention combining prediction with feedback could potentially optimise cognitive offloading in everyday life.

This paper presents a register-transfer-level (RTL) fault injection study of the LEON3 processor’s internal memory subsystem under single-event upsets (SEUs). The analysis targets four key components: the instruction cache (I-cache), data cache (D-cache), AHB bus control interface, and memory controller (MCTRL), all of which are unprotected in the standard LEON3 configuration. Using the NETFI+ fault injection framework, multi-cycle SEUs are injected into sequential elements across these blocks while executing a memory-intensive benchmark. The results show that the AHB interface is extremely fragile, with every fault causing execution failure. The memory controller, though architecturally invisible, frequently induces precise SPARC V8 traps such as window overflow and illegal instruction through indirect data-path corruption. The data cache is identified as the primary source of silent data corruption (SDC), while the instruction cache exhibits partial natural masking but remains susceptible to control-flow errors. These findings highlight the disproportionate impact of unprotected protocol and controller logic on system reliability and inform targeted hardening strategies for LEON3-based embedded systems in radiation-prone environments.

Outside the laboratory, people tend not to push working memory to its limits. Instead, we tend to capitalize on stable, external resources (e.g., assembly diagrams or shopping lists) in a dynamic, context-dependent trade-off with internal memory: we sample the environment more when remembering is "costly" (e.g., when a shopping list is difficult to remember) and remember more when sampling is costly (e.g. when the list is difficult to access). Here, we used our gamified Shopping Game paradigm to characterize this sampling-remembering trade-off in preschoolers (the youngest age at which it has been tested). In two preregistered experiments, 157 children (Exp. 1: 82 4.5-7-year-olds; Exp. 2: 75 5.5-7-year-olds, from the Northeastern US) used a touchscreen tablet to pick items from a virtual store based on a shopping list. Children could not see the list and store simultaneously, but could toggle between them. When we introduced a cost to access the list (a 4s lag before it appeared), children accessed it less, instead opting to remember more on each trip (but still less than their maximal capacity). In Experiment 1, we determined that children begin to respond to these costs at around 5.8 years of age. In Experiment 2, we showed that children exhausted the contents of their memory on each of their visits to the store: when it's up to them, young children seem to only load up their memory with as much as they intend to use. A video abstract of this article can be viewed at https://youtu.be/ZfJkod0Vj3k. SUMMARY: Here, we used our novel, gamified 'Shopping Game' paradigm to measure naturalistic memory use in 4.5-7-year-old children. Crucially, we could contrast how much a child could remember, how much they chose to remember, and how much of what was remembered was applied. We found that children will take advantage of an external resource (a shopping list) more and remember less, as the cost to access them decreases. This cost-dependent trade-off between sampling and remembering emerges around 5.8 years of age in our task.

Memorized performance is widely recognized as a a widely used marker of instrumental expertise because it releases performers’ attention from continuous note retrieval and allows greater focus on structural awareness, technical control, and expressive interpretation. Despite its importance, many students experience unstable memorization: recall may collapse under performance pressure, reliance on the written score remains strong, and practice frequently centers on mechanical repetition rather than deliberate encoding or reflective monitoring. This paper argues that such difficulties are not simply the result of limited memory capacity but emerge from the combined absence of a coherent internal memory structure and an effective regulatory approach to learning. To address this conceptual problem, the present paper examines two influential theoretical perspectives. Chaffin’s Performance Cues (PCs) theory explains how musical memory becomes organized through networks of meaningful reference points that connect semantic understanding, motor coordination, structural awareness, emotional intention, and auditory imagery. In parallel, Zimmerman’s Self-Regulated Learning (SRL) model conceptualizes learning as a cyclical process in which learners set goals, apply strategies, monitor performance, and engage in reflective evaluation supported by motivational beliefs such as self-efficacy and task value. Rather than treating these perspectives as separate lines of inquiry, this paper develops a theoretical analysis of their complementarity as well as their points of tension, including differing assumptions about temporal development, the sequencing of learning strategies, and the role of motivation within practice. Building on this analysis, the paper proposes an integrated conceptual framework in which SRL processes operate as regulatory pathways through which cue-based memory structures are intentionally constructed and stabilized. From this perspective, stable memorized performance emerges from the dynamic interaction between structural encoding and cyclical self-regulation rather than from repetition alone. The framework highlights pedagogical possibilities for guiding learners toward autonomous memorization by combining cue construction with explicit regulatory support, and it outlines directions for future theoretical and empirical work on the co-development of memory structure and learning regulation in instrumental practice.

This study presents a novel state-dependent fractional-order financial model that advances the modeling of memory effects in dynamic economic systems. By allowing the memory order q (t) to adapt continuously as a function of the internal state variable—specifically, the investment demand y (t)—the model captures the essential nonlinear feedback between market sentiment, memory depth, and macroeconomic behavior. The fractional framework is based on the Caputo-Fabrizio operator, chosen for its smooth, non-singular kernel, which ensures physical consistency and superior numerical tractability. A hyperbolic tangent structure is employed to define the adaptive memory function, enabling bounded, smooth, and symmetric evolution of q (t) within a realistic economic range. This design reflects empirically observed financial phenomena: heightened responsiveness during periods of uncertainty (short memory) and persistent inertia during stable or optimistic regimes (long memory). Numerical simulations demonstrate the model’s ability to replicate key financial features—such as damped oscillations, delayed stabilization, and variable sensitivity—under changing market conditions. Furthermore, the model is validated against real data from the 2008 global financial crisis, illustrating its empirical relevance and practical forecasting potential. By integrating internal memory regulation into the core of financial system dynamics, this work contributes a flexible and realistic tool for modern economic analysis, with applications in policy modeling, risk evaluation, and adaptive control of financial systems.

Older adults have impaired episodic memory abilities, but they can remember high-value information just as well as young adults and exhibit improved performance on memory-based tasks via cognitive offloading. For young adults, benefits from offloading a subset of memoranda (i.e., partial offloading) stem from both using the external memory aid to access offloaded information and better memory for nonoffloaded information, termed the saving-enhanced memory effect. Whether older adults also exhibit a saving-enhanced memory benefit from offloading is not yet known. The present study investigated if and how young and older adults' partial cognitive offloading behaviors and the benefits conferred by partial offloading change following experience with this strategy. Across two experiments, participants studied lists of words associated with varying point values under both internal memory and partial offloading conditions with the goal of earning as many points as possible on a subsequent free recall test. Participants chose a subset of words to offload before and after receiving three trials of direct instruction (Experiment 1) or extended practice (Experiment 2) using partial offloading. Across experiments, experience with partial offloading improved overall performance for both young and older adults. However, even after acquiring experience using partial offloading, young adults, but not older adults, exhibited better memory for nonsaved items, akin to the saving-enhanced memory effect. Thus, older adults benefitted from the use of an external memory aid, but internal memory resources freed up by offloading were not effectively rededicated to remembering nonoffloaded information as has been observed in young adults. (PsycInfo Database Record (c) 2026 APA, all rights reserved).

The fragmented Cioranian text represents an atypical discourse situated between literature, philosophy, and translation studies. The fragments function as autonomous yet interdependent units, requiring an interpretative rather than linear reading. Coherence relies not on narrative continuity but on subtle semantic relations such as equivalence, inclusion, and intersection. Repetition, coreference, and contextual contiguities create an internal textual memory, while the absence of explicit connectors forces the reader to reconstruct relationships through logic and cultural knowledge. As a result, the translator must convey not only lexemes but also the tension, ambiguity, and fragmentary structure. Semantic macrostructures and titles both divide and unify the discourse, generating coherence through thematic identity and variation. Rhetorical questions, paradoxes, and aphorisms establish a tacit dialogue with the reader. Part-whole relations shape meaning and require interpreting cultural presuppositions, while the visual iconicity of the fragment must be preserved. Thus, Cioran's fragmented text becomes a laboratory of implicit coherence that demands creative hermeneutics.

Large Language Models (LLMs) have evolved from simple n-gram systems to modern universal architectures; however, a key limitation remains the quadratic complexity of the self-attention mechanism with respect to input sequence length. This significantly increases memory consumption and computational costs, and with the emergence of tasks requiring extremely long contexts, creates the need for new architectural solutions. Since evaluating a proposed architecture typically requires long and expensive full-scale training, it is necessary to develop a tool that allows for a rapid preliminary assessment of a model’s internal memory capacity. This paper presents a method for quantitative evaluation of the internal memory of neural network architectures based on synthetic tests that do not require large data corpora. Internal memory is defined as the amount of information a model can reproduce without direct access to its original inputs. To validate the approach, a software framework was developed and tested on the GPT-2 and Mamba architectures. The experiments employed copy, inversion, and associative retrieval tasks. Comparison of prediction accuracy, error distribution, and computational cost enables a fast assessment of the efficiency and potential of various LLM architectures.

ABSTRACT In today’s digital world, people are documenting their lives more extensively than ever before. To investigate how this pervasive (digital) documentation shapes the way individuals reconstruct and recall personally relevant events, we conducted a preregistered experimental think-aloud study in which participants (N = 40; German sample) were asked to remember their birthdays from 2019 and 2024 in as much detail as possible. Participants completed the study in their usual home environments and were allowed to consult any external resources that they wanted to consult. The results demonstrate that participants almost exclusively used digital external resources. Moreover, participants relied more heavily on external resources when recalling the more distant birthday. Importantly, the use of external resources was an overall adaptive strategy, insofar as it helped participants gain new insights that went beyond what they could recall from internal memory alone. This provides further evidence that integrating information stored in one’s mind and information stored in the environment is a potentially beneficial and symbiotic process.

This paper presents a Liquid-Augmented Model Predictive Control (LA-MPC) framework for robust and adaptive motion control of quadrupedal robots operating under dynamic disturbances. The proposed approach integrates liquid neural dynamics into the predictive control loop, endowing the controller with real-time disturbance learning and model adaptation capabilities. System dynamics are formulated by linearizing single-rigid-body motion in three-dimensional space, while the liquid module continuously refines latent representations of unmodeled perturbations through its internal memory dynamics. The resulting hybrid predictive controller captures both short-term physical consistency and long-term disturbance evolution. By embedding the learned disturbance model within the MPC cost and constraint structure, the control law is reformulated as a quadratic program that can be solved efficiently in real time. Simulation on a quadrupedal platform demonstrates that the proposed LA-MPC achieves superior disturbance rejection, gait stability, and trajectory tracking accuracy compared to several popular learning baselines. The framework was further tested on the MuJoCo simulation platform, confirming its feasibility and practicality for agile quadrupedal locomotion in uncertain environments.

While Deep Neural Networks (DNNs) have achieved remarkable progress in Image Super-Resolution (SR) task, they face significant challenges for edge processing FHD images. Complex DNN operators lead to high hardware resource consumption and latency. Computational inefficiency of FPU increases energy consumption, while DDR access overhead and on-chip memory overflow further constrain real-time capabilities. To address this, we propose ISRLUT, a novel accelerator architecture focused on integer-only inference and near-memory computing. Its core contributions include: (1) Fusion of Neural LUT arithmetic with reconfigurable compute units, transforming unified LUT operators from DNN operators and enhancing hardware utilization; (2) An integer-only inference and parallel architecture, eliminating floating-point dependencies and significantly reducing energy consumption; (3) An innovative internal operator memory management scheme coupled with Tile-based Buffer Overlap and Private Cache Mechanism. We deploy ISRLUT on FPGA and ASIC platforms. Experiments demonstrate that ISRLUT achieves efficient performance: For 4 \(\times\) upscaling, it requires only 36.9 KB of storage and achieves a PSNR of 30.21 dB on Set5. Hardware implementation using a 55 nm ASIC consumes merely 0.0337 W power, delivers an energy efficiency of 7278.6 Mpixels/s/W, and achieves a real-time frame rate of 118 FPS for 4 \(\times\) FHD processing, validating its superiority in energy efficiency and hardware utilization.

Spatial cognitive behavior of pigeons induced by electrical stimulation in specific marked area.

Modern industrial automation systems are no longer restricted to proprietary Programmable Logic Controllers (PLC). They can be designed, developed, and commissioned on inexpensive yet powerful open-source microcontroller hardware. Open-source control software and communication protocols used in such systems provide increased protection against cyberattacks compared to proprietary controllers.1,2,3 OpenPLC1 is an open-source programming environment developed by Autonomy Logic that provides a platform for deploying industrial automation codes on hardware from different vendors. This work demonstrates the testing and benchmarking of OpenPLC functional blocks, ladder diagrams, etc., on common commercially available microcontroller boards and particularly on the indigenously developed C-DAC ARIES v2 microcontroller board with the THEJAS32 VEGA processor as a PLC. The RISC-V Processor Core and instruction set that run on the VEGA processor support a wide range of applications, including AI-based real-time process automation solutions.4 This work leads to the establishment of a solution framework powered by modular hardware with support for software that uses an extendable RISC-V Instruction Set to build more optimized industrial control system applications5,6,7 using Ladder Logic. This work demonstrates complex tasks such as fast data movement within internal memory, matrix multiplications, and crypto-computations in real-time on the microcontroller hardware, thus conveying that the solution is even capable of hosting an edge-level AI system for modern-day industrial applications. The tasks involved include a benchmarking algorithm that is executed on a candidate microcontroller hardware board. In addition to bare benchmarking using a C-code emulator for the respective boards, an extended version of the benchmarking was also carried out by encapsulating the benchmark algorithm as a PLC Ladder block. The benchmarking block helps control system designers select the right microcontroller unit for automation controllers at the field level, considering the availability of boards, cost, performance, security, and ease of integration.

The prediction of key quality variables plays an important role in industrial status identification and monitoring. Due to process disturbance and hard device limitation, data collection in modern industries often exhibits high noise and irregular data sampling. To solve the above problems, this article proposes a stacked supervised and reconstructed input denoising autoencoder integrated with internal attention long short-term memory (SSRDAE-IALSTM) network for soft sensing modeling. First, a stacked supervised and reconstructed input denoising autoencoder (SSRDAE) is designed. Compared with the original DAE, each supervised and reconstructed input DAE (SRDAE) can simultaneously reconstruct the process data and quality data at the output layer, aiming to reduce information loss and extract quality-related features. Second, the denoised features are fed into the interval attention LSTM (IALSTM) to adjust the influence of different historical samples on the current sample in irregular sampling data to capture long-term temporal features. Finally, performance validations are carried out on an industrial debutanizer column and a penicillin fermentation process. The experimental results show that the proposed model can enhance the learning ability of process features and obtain better prediction performance than other comparison methods.

Seismic station GIOL is the third station from the Bulgarian University Seismic Network, operated at the Faculty of Physics of Sofia University “St. Kliment Ohridski”. The station was recently installed (August 26th, 2025) and is equipped with the broadband seismometer SS08, and digitizer with internal memory SL06. Preliminary analyses of the registered seismograms include noise spectra analysis, seismic phase identification, earthquake spectra parameter determination, frequency-time analysis for dispersion curve extraction. Noise analysis reveals the station GIOL as the quietest one among all three stations. Seismic phase identification guided by theoretical travel times and joint analysis of records from several stations is necessary. Accumulation of a large number of earthquake spectra in time will lead to seismic moment magnitude definition. Dispersion curves indicate the properties of the structure along the wave propagation path. Results of the analyses are organized in a database facilitating future studies.

In older age, the ability to learn and remember new information is influenced by a myriad of factors, including the use of internal memory strategies. It has been proposed that individual differences in cognitive reserve (CR) may be related to older adults' spontaneous use of these strategies. The current study examined the relative importance and moderating effects of two CR proxies-educational attainment and crystallized intelligence-on older adults' use of a highly effective memory strategy, semantic clustering-grouping words together based on semantic relationships-during a verbal list-learning task. This study analyzed data from an archival sample of 185 older adults (n=83 with normal cognition, n=102 with mild cognitive impairment [MCI]) referred for neuropsychological assessment at a geriatric hospital in Ontario, Canada. A series of hierarchical regression models and relative weight analyses were conducted to examine the effects of education and crystallized intelligence (WASI Vocabulary scores) on semantic clustering, immediate recall, and delayed recall performance on the Kaplan-Baycrest Neurocognitive Assessment word lists subtest. A moderation analysis examined whether the relationship between semantic clustering and delayed recall was moderated by CR proxies. Gender and English language background predicted semantic clustering. Semantic clustering strategy-use accounted for additional variance in memory performance beyond the effects of demographic or reserve variables. While CR proxies did not significantly enhance the predictive value of any model, a moderation analysis found that crystallized intelligence was negatively associated with delayed recall performance in patients with MCI. Semantic clustering predicted delayed recall performance, and the effect was not moderated by education or crystallized intelligence. In a clinical sample of older adults presenting for neuropsychological assessment, women and native English speakers were more likely to use semantic clustering, independent of CR. The relationship between CR and memory performance is complex in a clinical setting, where patients may be assessed at different stages of disease progression. Semantic clustering appears to bolster memory performance regardless of an individual's existing level of CR, suggesting that teaching older adults with both low and high CR internal memory strategies may help to reduce everyday memory problems.

I want to know why this feels so familiar: Familiarity-detection during recall failure prompts curiosity and information seeking.

Temporal coherence-persistent alignment across time-can arise between agents with fundamentally distinct dynamics, yet classical diffusion models (Brownian motion, fractional Brownian motion, generalized Langevin equationswith shared noise) struggle under strong heterogeneity and asymmetry. We introduce the coupled memory graph process (CMGP), in which internal memory and directed, distance-gated coupling jointly produce synchronized behavior without reciprocity or common noise. Crucially, CMGP exhibits long-time coherence that reaches far beyond typical inherent memory times: an active particle with long-range memory remains temporally coherent with a subdiffusive partner despite mismatched scaling exponents. We show that this persistence arises from emergent long-range correlations generated by the coupling field rather than direct kernel overlap. Using Bayesian optimization, we identify broad parameter regions that support this "ghost coherence" (coherence without trajectory convergence) while preserving distinct exponents. These results outline a minimal mechanism for coordination in heterogeneous active systems and viscoelastic environments-one that standard stochastic models do not capture under comparable asymmetry unless augmented with explicit common drives or symmetric couplings.

A fundamental question in cognitive science is how information from internal memory is combined with external sensory input when making decisions. We hypothesized that previously learned and currently perceived information trade off against each other, such that extracting information from one source reduces the gathering and usage of information from the other. To test this hypothesis, we designed a two-armed bandit task where each arm is composed of both learned and perceived elements. We monitored participants’ gathering of perceptual information using eye tracking. Participants’ choices and gaze deployment showed a trade-off between the impact of learned and perceived information. The more a participant utilized internally stored learned information, the less they gathered perceptual information, and vice versa. Modeling participants’ information gathering indicated that the trade-off results from the faster gathering of learned information, which, when used, makes it less valuable to further invest effort in gathering additional perceptual information. Preliminary findings also suggested that an individual’s tendency to primarily rely on one source of information is a stable individual trait. These findings reveal how humans balance between learning and perception in forming decisions.