Liquid AI 发布 LongevityBench：17 项衰老生物学任务评测 18 个前沿 AI 系统

09.17.202609.17.2026An open benchmark and language models for AI in aging biologyBy Alex Zhavoronkov, Vladimir Naumov, Denis Sidorenko, Alex Aliper, Vladimir Aladinskiy, Ramin Hasani, Alexander Amini, Katerina Nasto, Mathieu Reymond, Rim Shayakhmetov, Zulfat Miftakhutdinov, and 2 othersBy Alex Zhavoronkov, Vladimir Naumov, Denis Sidorenko, Alex Aliper, Vladimir Aladinskiy, Ramin Hasani, Alexander Amini, Katerina Nasto, Mathieu Reymond, Rim Shayakhmetov, Zulfat Miftakhutdinov, Vadim N. Gladyshev, and Fedor GalkinShow more authorsWe introduce LongevityBench, an open suite of 17 tasks spanning five biodata domains, and use it to assess 18 frontier AI systems from six developer teams; 07.16.202607.16.2026In-Place Tokenizer Expansion for Pre-trained LLMsBy Jimmy T.H. Smith, Tarek Dakhran, Alberto Cabrera, Simon S. Lee, Paul Pak, Aditya Tadimeti, Tim Seyde, Maxime Labonne, Alexander Amini, and Mathias LechnerNew recipe expands pre-trained LLM tokenizers in place, cutting non-English token counts up to 4x with no quality loss; 06.16.202606.16.2026Fixed-Point Reasoners: Stable and Adaptive Deep Looped TransformersBy Sajad Movahedi, Vera Milovanović, Shlomo Libo Feigin, Alexander Theus, Thomas Hofmann, Valentina Boeva, T. Konstantin Rusch, and Antonio OrvietoFPRM is a Transformer-based Fixed-Point Reasoning Model for looped architectures, using pre-norm layers, residual scaling, and fixed-point halting to improve signal propagation, adapt compute to task difficulty, and deliver strong results on Sudoku, Maze, state-tracking, and ARC-AGI; 05.31.202605.31.2026Neural Low-Discrepancy SequencesBy Michael Etienne Van Huffel, Nathan Kirk, Makram Chahine, Daniela Rus, and T. Konstantin RuschNeuroLDS introduces the first machine learning framework for finite low-discrepancy sequences, generating point sequences with low prefix discrepancy and outperforming classical LDS methods across numerical integration, robot motion planning, scientific machine learning, and simulation; 05.29.202605.29.2026SwitchCraft: A Programmatic Framework for Designing State-Switching ProteinsBy Bowen Jing, Mihir Bafna, Anisha Parsan, Heyuan Michael Ni, David Kwabi-Addo, Bryan Bryson, Adam Klivans, and Bonnie BergerSwitchCraft is a programmable framework for designing state-switching proteins using backpropagation through compositional constraints and structure prediction models, enabling allosteric regulation, ligand discrimination, and de novo fluorescent biosensor design for biotechnology; 05.20.202605.20.2026MeMo: Model as a MemoryBy Ryan Wei Heng Quek, Sanghyuk Lee, Alfred Wei Lun Leong, Arun Verma, Alok Prakash, Nancy F. Chen, Bryan Kian Hsiang Low, Daniela Rus, and Armando Solar-LezamaMeMo (Memory as a Model) is a modular framework that adds timely, domain-specific knowledge to LLMs without changing model weights, capturing cross-document relationships, resisting retrieval noise, avoiding catastrophic forgetting, and supporting plug-and-play use with open and closed-source models; 04.22.202604.22.2026Preconditioned DeltaNet: Curvature-aware Sequence Modeling for Linear RecurrencesBy Neehal Tumma, Noel Loo, and Daniela RusPreconditioned delta-rule recurrences improve subquadratic long-context modeling by incorporating curvature from online least squares. The method links linear attention and the delta rule, introduces efficient variants of DeltaNet, GDN, and KDA, and boosts recall and language modeling performance; 03.26.202603.26.2026The Key to State Reduction in Linear Attention: A Rank-based PerspectiveBy Philipp Nazari and T. Konstantin RuschLinear attention offers a computationally efficient yet expressive alternative to softmax attention, maintaining a recurrent state that functions as a linear associative memory. However, recent empirical results indicate that the associative memory of trained linear attention models often exhibits a low-rank structure, suggesting that these models underexploit their capacity in practice. To illuminate this phenomenon, we provide a theoretical analysis of the role of rank in linear attention, revealing that low effective rank can affect retrieval error by amplifying query noise, as well as poorly condition query gradients; 03.02.202603.02.2026Low-Pass Flow MatchingBy Francesco M. Ruscio and T. Konstantin RuschFlow Matching typically relies on white noise sources, a choice often misaligned with the power spectra of natural data, which tend to decay with frequency. To address this, we introduce , a variant of Flow Matching based on an operator-modulated interpolant. This formulation induces a time-varying spectral bias that transitions from the source spectrum to a frequency-decaying bias as the path approaches the data. We validate our method on unconditional image generation tasks, including the scientific Galaxy10 dataset. Empirically, we show that our method is particularly effective when paired with adaptive ODE solvers, where it improves or preserves sample quality while substantially reducing sampling cost compared to standard baselines; 03.02.202603.02.2026AlphaQ: Calibration-Free Bit Allocation for Mixture-of-Experts QuantizationBy Wanqi Yang, Yuexiao Ma, Alexander Conzelmann, Xiawu Zheng, Michael W. Mahoney, T. Konstantin Rusch, and Shiwei LiuMixture-of-Experts (MoE) architectures scale computation via sparse expert activations, yet they remain memory-bound because all expert weights must reside in memory. Mixed-precision quantization can substantially reduce this footprint, but existing quantization methods estimate expert importance and assign bits based on calibration data. For frontier MoE LLMs, however, the original training data (and thus its true training distribution) is proprietary and inaccessible. Thus, any calibration set is at best a surrogate and may yield a biased or incomplete view of expert utilization, leading to suboptimal bit allocation. To address these problems, we propose AlphaQ, a novel calibration-free bit-allocation method for MoE quantization; 01.26.202601.26.2026Zero-Overhead Introspection for Adaptive Test-Time ComputeBy Rohin Manvi, Joey Hong, Tim Seyde, Maxime Labonne, Mathias Lechner, and Sergey LevineLarge language models excel at reasoning but lack key aspects of introspection, including the ability to anticipate their own success and the computation required to achieve it. Humans use real-time introspection to decide how much effort to invest, when to make multiple attempts, when to stop, and when to signal success or failure. Without this ability, LLMs struggle to make intelligent meta-cognition decisions. Test-time scaling methods such as Best-of-N drive up cost and latency by using a fixed budget of samples regardless of the marginal benefit of each one at any point in generation, and the absence of confidence signals can mislead people, prevent appropriate escalation to better tools, and undermine trustworthiness; 01.26.202601.26.2026Learning residue level protein dynamics with multiscale GaussiansBy Mihir Bafna, Bowen Jing, and Bonnie BergerMany methods have been developed to predict static protein structures, however understanding the dynamics of protein structure is essential for elucidating biological function. While molecular dynamics (MD) simulations remain the in silico gold standard, its high computational cost limits scalability. We present DynaProt, a lightweight, SE(3)-invariant framework that predicts rich descriptors of protein dynamics directly from static structures; 01.26.202601.26.2026Antislop: A Comprehensive Framework for Identifying and Eliminating Repetitive Patterns in Language ModelsBy Samuel J Paech, Allen G Roush, Judah Goldfeder, and Ravid Shwartz-ZivAntislop detects and removes repetitive LLM “slop” using inference-time suppression, profiling, and targeted fine-tuning; 01.26.202601.26.2026The Curious Case of In-Training Compression of State Space ModelsBy Makram Chahine, Philipp Nazari, Daniela Rus, and T. Konstantin RuschState Space Models (SSMs), developed to tackle long sequence modeling tasks efficiently, offer both parallelizable training and fast inference. At their core are recurrent dynamical systems that maintain a hidden state, with update costs scaling with the state dimension. A key design challenge is striking the right balance between maximizing expressivity and limiting this computational burden. Control theory, and more specifically Hankel singular value analysis, provides a potent framework for the measure of energy for each state, as well as the balanced truncation of the original system down to a smaller representation with performance guarantees; 12.01.202512.01.2025LFM2 Technical ReportBy Alexander Amini, Anna Banaszak, Harold Benoit, Arthur Böök, Tarek Dakhran, Song Duong, Alfred Eng, Fernando Fernandes, Marc Härkönen, Anne Harrington, Ramin Hasani, and 22 othersBy Alexander Amini, Anna Banaszak, Harold Benoit, Arthur Böök, Tarek Dakhran, Song Duong, Alfred Eng, Fernando Fernandes, Marc Härkönen, Anne Harrington, Ramin Hasani, Saniya Karwa, Yuri Khrustalev, Maxime Labonne, Mathias Lechner, Valentine Lechner, Simon Lee, Zetian Li, Noel Loo, Jacob Marks, Edoardo Mosca, Samuel J. Paech, Paul Pak, Rom N. Parnichkun, Alex Quach, Ryan Rogers, Daniela Rus, Nayan Saxena, Bettina Schlager, Tim Seyde, Jimmy T.H. Smith, Aditya Tadimeti, and Neehal TummaShow more authorsWe present LFM2, a family of Liquid Foundation Models designed for efficient on-device deployment and strong task capabilities. Using hardware-in-the-loop architecture search under edge latency and memory constraints, we obtain a compact hybrid backbone that combines gated short convolutions with a small number of grouped query attention blocks, delivering up to 2x faster prefill and decode on CPUs compared to similarly sized models; 09.25.202509.25.2025RL Squeezes, SFT Expands: A Comparative Study of Reasoning LLMsBy Kohsei Matsutani, Shota Takashiro, Gouki Minegishi, Takeshi Kojima, Yusuke Iwasawa, and Yutaka MatsuoLarge language models (LLMs) are typically trained by reinforcement learning (RL) with verifiable rewards (RLVR) and supervised fine-tuning (SFT) on reasoning traces to improve their reasoning abilities. However, how these methods shape reasoning capabilities remains largely elusive. Going beyond an accuracy-based investigation of how these two components sculpt the reasoning process, this paper introduces a novel analysis framework that quantifies reasoning paths and captures their qualitative changes under each training process (with models of 1.5B, 7B, and 14B parameters on mathematical domains); 06.06.202506.06.2025Exploring Diffusion Transformer Designs via GraftingBy Keshigeyan Chandrasegaran, Michael Poli, Daniel Y. Fu, Dongjun Kim, Lea M. Hadzic, Manling Li, Agrim Gupta, Stefano Massaroli, Azalia Mirhoseini, Juan Carlos Niebles, Stefano Ermon, and 1 otherBy Keshigeyan Chandrasegaran, Michael Poli, Daniel Y. Fu, Dongjun Kim, Lea M. Hadzic, Manling Li, Agrim Gupta, Stefano Massaroli, Azalia Mirhoseini, Juan Carlos Niebles, Stefano Ermon, and Li Fei-FeiShow more authorsGrafting edits pretrained diffusion transformers into efficient hybrids using <2% compute, delivering strong quality and up to 1.43x speedups; 04.28.202504.28.2025Quantifying Memory Utilization with Effective State-SizeBy Rom N. Parnichkun, Neehal Tumma, Armin W. Thomas, Alessandro Moro, Qi An, Taiji Suzuki, Atsushi Yamashita, Michael Poli, and Stefano MassaroliIntroducing effective state-size (ESS), a metric that quantifies memory utilization in sequence models to improve initialization and distillation; 11.26.202411.26.2024STAR: Synthesis of Tailored ArchitecturesBy Armin W. Thomas, Rom Parnichkun, Alexander Amini, Stefano Massaroli, and Michael PoliSTAR uses evolutionary search over architecture genomes to automatically design hybrids that beat Transformers on quality, size, and cache; 07.26.202407.26.2024Towards a theory of learning dynamics in deep state space modelsBy Jakub Smekal, Jimmy Smith, Michael Kleinman, Dan Biderman, and Scott LindermanTowards a theory of learning dynamics in deep state space models; 06.16.202406.16.2024The Empirical Impact of Neural Parameter Symmetries, or Lack ThereofBy Derek Lim, Theo Putterman, Robin Walters, Haggai Maron, and Stefanie JegelkaThe Empirical Impact of Neural Parameter Symmetries, or Lack Thereof; 05.01.202405.01.2024Large Scale Dataset Distillation with Domain ShiftBy Noel Loo, Alaa Maalouf, Ramin Hasani, Mathias Lechner, Alexander Amini, and Daniela RusLarge Scale Dataset Distillation with Domain Shift; 05.01.202405.01.2024LLM and Simulation as Bilevel Optimizers: A New Paradigm to Advance Physical Scientific DiscoveryBy Pingchuan Ma, Tsun-Hsuan Wang, Minghao Guo, Zhiqing Sun, Joshua B. Tenenbaum, Daniela Rus, Chuang Gan, and Wojciech MatusikLLM and Simulation as Bilevel Optimizers: A New Paradigm to Advance Physical Scientific Discovery; 05.01.202405.01.2024RoboGen: Towards Unleashing Infinite Data for Automated Robot Learning via Generative SimulationBy Yufei Wang, Zhou Xian, Feng Chen, Tsun-Hsuan Wang, Yian Wang, Katerina Fragkiadaki, Zackory Erickson, David Held, and Chuang GanRoboGen: Towards Unleashing Infinite Data for Automated Robot Learning via Generative Simulation; 05.01.202405.01.2024Position: Future Directions in the Theory of Graph Machine LearningBy Christopher Morris, Fabrizio Frasca, Nadav Dym, Haggai Maron, Ismail Ilkan Ceylan, Ron Levie, Derek Lim, Michael M. Bronstein, Martin Grohe, and Stefanie JegelkaPosition: Future Directions in the Theory of Graph Machine Learning; 05.01.202405.01.2024Mechanistic Design and Scaling of Hybrid ArchitecturesBy Michael Poli, Armin W Thomas, Eric Nguyen, Pragaash Ponnusamy, Björn Deiseroth, Kristian Kersting, Taiji Suzuki, Brian Hie, Stefano Ermon, Christopher Re, Ce Zhang, and 1 otherBy Michael Poli, Armin W Thomas, Eric Nguyen, Pragaash Ponnusamy, Björn Deiseroth, Kristian Kersting, Taiji Suzuki, Brian Hie, Stefano Ermon, Christopher Re, Ce Zhang, and Stefano MassaroliShow more authorsMechanistic Design and Scaling of Hybrid Architectures; 05.01.202405.01.2024State-Free Inference of State-Space Models: The *Transfer Function* ApproachBy Rom Parnichkun, Stefano Massaroli, Alessandro Moro, Jimmy T.H. Smith, Ramin Hasani, Mathias Lechner, Qi An, Christopher Re, Hajime Asama, Stefano Ermon, Taiji Suzuki, and 2 othersBy Rom Parnichkun, Stefano Massaroli, Alessandro Moro, Jimmy T.H. Smith, Ramin Hasani, Mathias Lechner, Qi An, Christopher Re, Hajime Asama, Stefano Ermon, Taiji Suzuki, Michael Poli, and Atsushi YamashitaShow more authorsState-Free Inference of State-Space Models: The *Transfer Function* Approach; 04.05.202404.05.2024Growing Q-Networks: Solving Continuous Control Tasks with Adaptive Control ResolutionBy Tim Seyde, Peter Werner, Wilko Schwarting, Markus Wulfmeier, and Daniela RusGrowing Q-Networks: Solving Continuous Control Tasks with Adaptive Control Resolution; 01.16.202401.16.2024Understanding Reconstruction Attacks with the Neural Tangent Kernel and Dataset DistillationBy Noel Loo, Ramin Hasani, Mathias Lechner, Alexander Amini, and Daniela RusUnderstanding Reconstruction Attacks with the Neural Tangent Kernel and Dataset Distillation; 01.16.202401.16.2024Graph Metanetworks for Processing Diverse Neural ArchitecturesBy Derek Lim, Haggai Maron, Marc T. Law, Jonathan Lorraine, and James LucasGraph Metanetworks for Processing Diverse Neural Architectures