scira
New threadAboutBlogTermsPrivacyGitHubFeedback
SupportSettings
Sign in to save threadsLibrary, Projects, Lookout and Apps need an account.Sign in
AcademicAcademic

Friday, Nov 20 · 43 million abstracts

New research, explained

Academic
AI & computer scienceMedicineNeuroscienceClimateGenetics
Academic

Sources

arXivPhysics, maths, CS
PubMedBiomedical literature
bioRxivBiology preprints
medRxivMedical preprints
Ask Scira
Main approaches to protein structure prediction since AlphaFoldPapers by Yann LeCun on self-supervised learning since 2023Does intermittent fasting improve insulin sensitivity?
Students$9/month for Scira Pro with a university emailGet student pricing

AI & computer science

Last 45 days
Last 45 days in AI & computer science

This week's papers explore advanced architectural paradigms for LLM agents to improve efficiency, reasoning robustness, and working memory management during complex, long-horizon tasks. The most notable result is the introduction of novel frameworks—such as dynamic model collaboration, hybrid latent-text reasoning, and proactive context self-allocation—that successfully mitigate cognitive overload and coordination trade-offs without relying on external routing modules.

arXiv2603.26034
AgentCollab: A Self-Evaluation-Driven Collaboration Paradigm for Efficient LLM Agents

Autonomous agents powered by large language models (LLMs) perform complex tasks through long-horizon reasoning and tool interaction, where a fundamental trade-off arises between execution efficiency and reasoning robustness. Models at different capability-cost levels offer complementary advantages: lower-cost models enable fast execution but may struggle on difficult reasoning segments, while stronger models provide more robust reasoning at higher computational cost. We present AgentCollab, a self-driven collaborative inference framework that dynamically coordinates models with different reasoning capacities during agent execution. Instead of relying on external routing modules, the framework uses the agent's own self-reflection signal to determine whether the current reasoning trajectory is making meaningful progress, and escalates control to a stronger reasoning tier only when necessary. To further stabilize long-horizon execution, we introduce a difficulty-aware cumulative escalation strategy that allocates additional reasoning budget based on recent failure signals. In our experiments, we instantiate this framework using a two-level small-large model setting. Experiments on diverse multi-step agent benchmarks show that AgentCollab consistently improves the accuracy-efficiency Pareto frontier of LLM agents. The code is available at https://github.com/SparkEngineAI/AgentFlashInfer.

Ask about this paper: AgentCollab: A Self-Evaluation-Driven Collaboration Paradigm for Efficient LLM Agents
arXiv2610.18861
Hybrid Reasoning Agents

How can we build an agent that thinks beyond human language, yet still acts through it in the external environment? Modern LLM agents still reason and interact through text, while latent reasoning offers an efficient internal medium but lacks native tool access and reasoning transparency. We bridge these two directions with Hybrid Reasoning Agents (HybridAgents), which dynamically navigate between latent reasoning and text-based interaction throughout a unified generation process. Architecturally, a lightweight HybridLink enables seamless transitions between text and latent modes, while the agent's continuous working memory preserves lossless hybrid context across multiple turns. These designs preserve tool-integrated agentic capabilities while making latent thoughts directly inspectable. Building on this architecture, we further develop dedicated post-training algorithms for hybrid reasoning: HybridSFT stabilizes tool-integrated hybrid generation by teaching the backbone to coordinate with HybridLink, then HybridRL scales reasoning by learning when to think in latent space and when to act through text. We evaluate HybridAgent at 3 model scales across 11 benchmarks covering mathematics, science, medicine, search, and agentic reasoning. Compared with frontier LLM Agents and agentic frameworks spanning both text-based and latent methods, HybridAgent consistently improves performance by 10.4%, while delivering 35.5%-54.7% token reduction, 1.8x-2.6x faster inference, and 2.7x-3.7x cost savings. Code and data are available at https://github.com/togethercomputer/HybridAgent.

Ask about this paper: Hybrid Reasoning Agents
arXiv2610.40350
Don't Lose the Thread: Empowering Long-Horizon LLM Agents with Cognitive Resource Self-Allocation

Agents powered by large language models (LLMs) have demonstrated remarkable progress in solving complex reasoning tasks. However, LLM agents often falter on long-horizon tasks due to cognitive overload, as their working memory becomes cluttered with expanding and irrelevant information, which dilutes their attention and hinders effective planning and reasoning. To mitigate this challenge, we introduce Cognitive Resource Self-Allocation (CORAL), a novel reasoning paradigm that empowers agents to proactively optimize their context. Implemented as an agent-callable working memory management toolset, CORAL allows an agent to checkpoint its task progress and verified facts, and adaptively initiate a new problem-solving episode by resetting cluttered context while preserving only the checkpointed information, enabling it to resume reasoning from a clean state. We further enhance the agent's checkpointing capabilities using a Multi-episode Agentic Reinforced Policy Optimization algorithm. On the GAIA benchmark, CORAL with an 8B model achieves 42.8% average accuracy, with particularly strong gains on the most challenging long-horizon tasks (Level 2 and Level 3). Crucially, these gains come at a substantially lower inference cost: CORAL consumes 6-7× fewer input tokens than a ReAct agent. It also generalizes to multi-hop Knowledge QA benchmarks and to the considerably harder BrowseComp, where it improves a strong proprietary agent by 8.0 points.

Ask about this paper: Don't Lose the Thread: Empowering Long-Horizon LLM Agents with Cognitive Resource Self-Allocation

Medicine

Last 45 days
Last 45 days in Medicine

This week's papers collectively explore the efficacy and clinical outcomes of various interventions—specifically Tai Chi, Baduanjin, thunder-fire moxibustion combined with tincture, and the oral JAK1/TYK2 inhibitor QY201—for managing conditions such as knee osteoarthritis, cancer-related fatigue, and atopic dermatitis. However, because the provided abstracts omit the results, the specific most notable outcome cannot be determined from the text.

PubMed CentralPMC13750279
An Exploratory Three-Arm Randomized Controlled Trial: The Clinical Effects of Tai Chi, Baduanjin, and Health Education in the Management of Knee Osteoarthritis.

Objective: This three-arm randomized controlled trial aimed to investigate the comparative clinical effects of Tai Chi (TC), Baduanjin(BDJ), and Health Education(HE) for controlling symptoms and restoring physical function in individuals diagnosed with knee osteoarthritis(KOA).…

Ask about this paper

Neuroscience

Last 45 days
Last 45 days in Neuroscience

This week's papers collectively explore the neural mechanisms, brain interactions, and stimulation interventions underlying memory and cognitive function in aging populations. The most notable result is the evaluation of non-invasive brain stimulation targeting temporal and parietal cortices as a promising approach for enhancing memory decline in healthy older adults, those with mild cognitive impairment, and patients with Alzheimer's disease.

PubMed43009411
Functional brain signatures of successful cognitive aging.

Cognitive aging trajectories vary markedly. While many older adults experience gradual decline in memory and executive function, some maintain high cognitive performance well into advanced age. Understanding the neural mechanisms underlying this variability carries important…

Ask about this paper: Functional brain signatures of successful cognitive aging.

Climate

Last 45 days
Last 45 days in Climate

This week's papers collectively examine climate change adaptation strategies across diverse vulnerable populations and urban contexts, synthesizing how nature-based solutions, everyday adjustments, and on-farm behaviors are implemented. The most notable result is the finding that while urban green spaces dominate documented nature-based solutions in African cities—appearing in nearly half of the analyzed studies—urban agriculture remains largely under-recognized, alongside the identification of distinct pathways that drive incremental versus transformative on-farm adaptations.

PubMed CentralPMC13681004
A systematic review of nature-based solutions for climate action in African cities.

Nature-based solutions (NbS), including ecosystem-based adaptation, are increasingly proposed for climate action in African cities, but the evidence base is fragmented and lacks systematic synthesis of what is being implemented, where, and how effectively. We analysed 112 studies retained from a systematic review, covering 134 urban areas across 42 countries and including 45 informal settlements. Five NbS typology classes are inductively derived including: Green spaces; Rivers, catchments and drainage systems; Coastal systems; Green buildings; and Urban agriculture and multi-use gardens. Urban green spaces dominate the evidence (49% of studies) and are the most multi-functional class. Urban agriculture and multi-use gardens are under-recognised in headline framings, named as primary in only 5% of studies but appearing in 34% of the wider intervention mix. Evidence is increasing but concentrates geographically on larger cities. Successful climate change adaptation and mitigation outcomes are reported by 72% and 27% of studies, respectively. Heat mitigation is the most frequently quantified adaptation outcome, with temperature reductions up to 12.2 °C reported in proximity to urban green and blue spaces across the continent. Qualitative evidence links urban NbS to effective coastal protection, flood risk reduction, and increased water security. Multi-stakeholder governance anchored by local actors, and meaningful community participation are consistent enablers of effective implementation. However, insufficient long-term funding and livelihood disruptions and displacement raise concerns about long-term impacts. Priority research gaps include longitudinal effectiveness monitoring, financing mechanisms for informal contexts, systematic assessment of mechanisms behind inequitable outcomes, and evidence from under-represented regions and smaller cities.

Genetics

Last 45 days
Last 45 days in Genetics

This week's papers collectively highlight the rapid expansion of CRISPR technologies—including base editors, prime editors, and RNA-targeting effectors—beyond standard Cas9 nucleases for diverse functional genomics and gene therapy applications. The most notable result is the clinical translation of these tools, evidenced by regulatory approval for ex vivo therapies treating hemoglobinopathies and the ongoing clinical advancement of in vivo and base-editing candidates for genetic disorders and cancers.

PubMed42771304
CRISPR/Cas Systems: Biological Basis and Genome Editing Applications.

Advances in genome editing technologies, including CRISPR-mediated knockout, base editing, and prime editing, are explored with an emphasis on their mechanisms and applications. The chapter also examines emerging CRISPR-based platforms for transcriptional regulation, epigenome…

Ask about this paper: CRISPR/Cas Systems: Biological Basis and Genome Editing Applications.
: An Exploratory Three-Arm Randomized Controlled Trial: The Clinical Effects of Tai Chi, Baduanjin, and Health Education in the Management of Knee Osteoarthritis.
PubMed43040124
Clinical outcomes of Thunder-fire moxibustion with or without Yiqi Wenyang Tincture for cancer-related fatigue: a randomized controlled trial.

BACKGROUND: Cancer-related fatigue (CRF) is a debilitating symptom with limited treatment options. This study evaluated whether thunder-fire moxibustion (TFM) combined with Yiqi Wenyang Tincture (YWT) could provide additional benefit for CRF beyond conventional integrative…

Ask about this paper: Clinical outcomes of Thunder-fire moxibustion with or without Yiqi Wenyang Tincture for cancer-related fatigue: a randomized controlled trial.
PubMed42901423
Efficacy and safety of QY201, an oral dual Janus kinase 1/tyrosine kinase 2 inhibitor, in adults with moderate-to-severe atopic dermatitis: a multicenter, randomized, double-blind, placebo-controlled phase 2 dose-finding trial.

METHODS: In this phase 2, randomized, double-blind, placebo-controlled trial, 200 adults with moderate-to-severe AD were randomized 1:1:1:1 to receive QY201 (5, 10, or 20 mg twice daily) or placebo for 12 weeks.

Ask about this paper: Efficacy and safety of QY201, an oral dual Janus kinase 1/tyrosine kinase 2 inhibitor, in adults with moderate-to-severe atopic dermatitis: a multicenter, randomized, double-blind, placebo-controlled phase 2 dose-finding trial.
PubMed CentralPMC13729879
Hippocampal-prefrontal cortical interactions and neural representations of memory.

The ability to store memories and access past experiences when making decisions in the present is fundamental to cognition. The hippocampus is critical for these processes and has been a primary site of investigation in the learning and memory field for the last several decades.…

Ask about this paper: Hippocampal-prefrontal cortical interactions and neural representations of memory.
PubMed43044296
Effects of temporal and parietal cortex non-invasive brain stimulation on memory in healthy older adults, mild cognitive impairment, and Alzheimer's disease: A systematic review and meta-analysis.

Memory decline is a common cognitive complaint among healthy older adults, mild cognitive impairment (MCI) and Alzheimer's disease (AD). Non-invasive brain stimulation (NIBS) has been investigated as a promising approach for enhancing memory. While most studies target the…

Ask about this paper: Effects of temporal and parietal cortex non-invasive brain stimulation on memory in healthy older adults, mild cognitive impairment, and Alzheimer's disease: A systematic review and meta-analysis.
Ask about this paper: A systematic review of nature-based solutions for climate action in African cities.
PubMed CentralPMC7619734
Living with extreme weather: Health vulnerabilities and everyday adaptations of informal outdoor workers in urban Vietnam.

Climate change has intensified health risks for informal outdoor workers in the Global South, whose exposure to extreme weather is compounded by systemic marginalisation, economic precarity, and limited access to healthcare. Yet dominant climate-health research and adaptation…

Ask about this paper: Living with extreme weather: Health vulnerabilities and everyday adaptations of informal outdoor workers in urban Vietnam.
PubMed43021021
Changing with the climate: A review of on-farm adaptation behaviour dynamics for research, practice, and policy.

Understanding behavioural dynamics behind farm-level climate adaptation is crucial for research, policy, and practice. Incremental adaptations rely heavily on the practical PBC-pathway, whereas transformative adaptations are more shock-driven and ideological, operating through…

Ask about this paper: Changing with the climate: A review of on-farm adaptation behaviour dynamics for research, practice, and policy.
PubMed CentralPMC13748939
CRISPR Technology: From Bench to Bedside Applications.

CRISPR-Cas genome editing represents one of the most transformative advances in modern molecular biology, enabling programmable site-specific genetic modification that has revolutionized functional genomics and precision therapeutic development. Beyond canonical Cas9 nucleases, the CRISPR toolkit has diversified into precision base editors, prime editors, CRISPR-associated transposases, and Cas12/Cas13 RNA-targeting effectors, each with distinct mechanistic properties. Clinically, ex vivo CRISPR therapies for hemoglobinopathies have secured regulatory approval, while base editing, prime editing, and in vivo Chimeric Antigen Receptor (CAR)-T candidates are advancing through trials for a broad spectrum of hematologic, oncologic, and rare genetic disorders. However, existing reviews largely address technical innovation or clinical translation in isolation, lacking a holistic, up-to-date synthesis of the full bench-to-bedside trajectory of CRISPR technology. This review systematically dissects the molecular mechanisms, advantages, and limitations of core CRISPR editing platforms; summarizes pivotal clinical data across hemoglobinopathies, CAR-T immunotherapy, muscular dystrophy, cystic fibrosis, cardiometabolic diseases and rare monogenic diseases; evaluates delivery systems including lipid nanoparticles and viral vectors; and discusses key translational barriers such as off-target genotoxicity and immunogenicity alongside emerging optimization strategies. This work provides a holistic bench-to-bedside framework for CRISPR research, offering critical insights to advance next-generation precision genetic medicine.

Ask about this paper: CRISPR Technology: From Bench to Bedside Applications.
PubMed42973349
CRISPR/Cas9-based Genome Editing in Inherited Diseases: Mechanisms, Therapeutic Advances, Safety and Clinical Translation.

Among genome-editing technologies, the CRISPR/Cas9 system has emerged as a versatile and efficient platform due to its programmability, simplicity, and precision. This review summarizes the evolution of gene therapy and highlights the therapeutic potential of CRISPR-based genome…

Ask about this paper: CRISPR/Cas9-based Genome Editing in Inherited Diseases: Mechanisms, Therapeutic Advances, Safety and Clinical Translation.