22 papers · ranked by Valyu relevance
Taosha Guo, Fabio Pasqualetti
The Hopfield network established that static memories can be stored as energy minima of a recurrent dynamical system, yet real intelligent agents must navigate \emph{sequences} of memories rather than isolated snapshots.Biological cortex addresses this through a separation of timescales: fast synaptic dynamics encode…
Songwei Dong, Zihan Chen, Chengshuai Shi, Peng Wang + 2 more
Memory plays a central role in enabling large language models (LLMs) to operate over sequential tasks by accumulating and reusing experience over time. However, existing evaluations of LLM memory mostly rely on aggregate metrics such as final hold-out accuracy or cumulative online performance, which can obscure…
Yiren Ren, Vishwadeep Ahluwalia, Claire Arthur, Thackery Brown
The human brain continuously segments experience into meaningful episodes while also encoding temporal relationships between events, yet the mechanisms that optimize this dual challenge remain poorly understood. Here we tested a theoretical framework in which structured temporal context from one modality (music) can…
Margot Wagner, Yusi Chen, Arjun Karuvally, Mia Cameron + 1 more
The hippocampus must balance stable memory representations with internally generated sequential dynamics underlying replay, prediction and traveling waves. How hippocampal circuitry achieves both remains unclear. Here, we show that recurrent neural networks trained on continuous prediction tasks converge to a…
Qiaoli Huang, Christian F. Doeller
Human cognition is capacity-limited, requiring strategies to actively structure information. Eye movements offer a natural mechanism for sequential sampling, but whether such sequences organize mnemonic representations is unknown. We developed a working-memory task where color-frequency pairings created a consistent…
Zikui Cai, Kaushal Janga, Tan Dat Dao, Seungjae Lee + 14 more
Embodied question answering (EQA) is traditionally evaluated under an episodic formulation, where agents solve each task independently and reset internal state between episodes. However, real-world robots operate continuously and must accumulate, retain, and selectively reuse information acquired from prior…
Zihan Chen, Songwei Dong, Chengshuai Shi, Peng Wang + 3 more
Sequentially evolving LLM memory enables agents to reuse past experience, but existing systems usually deploy each locally generated memory update without checking whether it improves future behavior. As a result, updates that help the current task may overwrite useful knowledge, introduce over-specific rules, or bias…
Ainsley Temudo, Owen Benzley, Bradley R. King, Genevieve Albouy
Everyday activities often require learning sequences that necessitate the involvement of both the declarative and the procedural memory domains. Previous research has shown that a learning structure that is common across tasks from different domains can improve learning and resistance to interference. However, it…
Eduardo A. Aponte, Thanneer M. Perumal, Francesca Cormack, Christopher H. Chatham + 1 more
The last decades have seen a great improvement in our understanding of visuospatial working memory (VSWM). Despite this progress, less is known about how information is stored, retained, and removed from VSWM when novel information is presented sequentially. Here, we present a novel computational model of the dynamics…
Tatsuya Haga
Hippocampus is known to replay activity patterns to recall and process memories, which is often related to Hopfield-type attractor dynamics. Another line of theoretical studies suggests that hippocampal replay prioritizes replay of experiences to accelerate value learning for efficient decision making. It is unknown…
Shujia Chen, Don Straney, Damiano Pasini
Recent advances in mechanical computing have harnessed bistable mechanisms with intrinsic memory to extend the scope of physically embodied intelligence, enabling history-dependent behavior. However, existing mechanical computing architectures largely fail to integrate mechanically encoded memory with the logic…
Qi Zhang
Declarative memory, or explicit memory, can be fractionated into episodic memory and semantic memory (Tulving, 1972). Episodic memory refers to personally based memories and semantic memory refers to the memory of factual knowledge. The loss of the capacity in retaining episodic memory leads to various amnesias, e.g.…
Zhang, Qi
Explicit (declarative) memory, the memory that can be "declared" in words or languages, is made up of two dissociated parts: episodic memory and semantic memory. This dissociation has its neuroanatomical basis—episodic memory is mostly associated with the hippocampus and semantic memory with the neocortex. The two…
Xu Pan, Ely Hahami, Roy Siegelmann, Haim Sompolinsky
Humans retain memories of individual experiences for a lifetime, an ability attributed to a complementary learning system in which a fast process encodes episodes and a slow process integrates them into semantic knowledge. In classical Hebbian models such as Hopfield networks, memory traces are superposed in shared…
Pengfei Sun, Zhe Su, Jascha Achterberg, Giacomo Indiveri + 2 more
Spiking neural networks excel at event-driven sensing. Yet, maintaining task-relevant context over long timescales both algorithmically and in hardware, while respecting both tight energy and memory budgets, remains a core challenge in the field. Here we address this challenge through an algorithm-hardware co-design…
Chong Zhao, Temilade Adekoya, Sintra Horwitz, Edward Awh + 1 more
Working memory (WM) tasks often require comparing remembered items to test displays, but little is known about how people selectively remove irrelevant information at test. Across three experiments, we used contralateral delay activity (CDA) to track WM load and examine selective removal. In Experiment 1, CDA…
Authors not listed
Accurate prediction of chemical reaction yields remains essential for accelerating synthesis optimization, yet current machine learning models face critical limitations in capturing temporal dynamics, providing calibrated uncertainty estimates, and explicitly modeling reactant-to-product transformations. Here we…
Maria Nemeth, Christian Frings, Birte Moeller
In theories of human action, it is assumed that individual actions are nested within higher-order action plans. This hierarchical structure oftentimes allows for the anticipatory planning of multiple future actions even before the current action is fully executed or situational cues demand this specific action.…
Alice Mason, Geoff Ward, Gordon D. A. Brown, Simon Farrell
We often need to make retrospective judgments about experiences, either to communicate our impression to others, or to provide some basis for choosing between options. While the best summative representation of an experience as a whole would usually be the sum of the individual elements of the experience, people’s…
Authors not listed
Machine olfaction—the artificial replication of the sense of smell—faces significant challenges due to the absence of large, standardized training datasets. Unlike vision, language, and audio models, which benefit from extensive corpora such as ImageNet, GLUE, and AudioSet, olfaction lacks scaled equivalents and…
Authors not listed
This paper develops a comprehensive theoretical framework for designing quantum memory systems with enhanced resilience to thermal decoherence through engineered lattice geometries and protective structures. We formulate a unified mathematical description connecting material properties, geometric configurations, and…
Authors not listed
A framework for catalysis based on categorical aperture selection rather than temporal acceleration is presented. Traditional catalysis theory describes catalysts as agents that accelerate reactions by lowering activation energies, implicitly treating time as the fundamental variable and reaction rate enhancement as…