paperlist.ai

AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

MonitrLLM: A Community-Centered Evaluation Infrastructure for Large Language Models

Victor Ojewale, Ro Encarnación, Suresh Venkatasubramanian, Danaé Metaxa

Benchmark suites assess model capability on controlled tasks; large-scale conversation corpora capture naturalistic use without user feedback; and in-interface feedback mechanisms record satisfaction without task purpose. Together, they leave a critical gap in LLM evaluation: no existing infrastructure routinely links interaction trajectories to user-defined outcomes. We introduce MonitrLLM, open-source infrastructure for community-centered LLM evaluations that links full conversation transcripts to user-reported task intent and outcome assessments, treating all three as primary evaluative signals rather than optional metadata. To demonstrate the value of this approach, we conducted a two-week feasibility pilot with 26 college students using ChatGPT, collecting 206 evaluation reports with full conversation transcripts. The findings from our pilot demonstrate the value of connecting conversation trajectories with user-reported outcomes. For instance, despite reporting high average satisfaction (4.19/5) with their LLM interactions, participants also experience a substantial 23.1% failure rate on their goal tasks. We also find that multi-turn conversations are reported as failing at 2.5 times the rate of single-turn exchanges, a pattern that reframes extended interaction as a signal of difficulty rather than engagement. We conclude by discussing the value of incorporating direct user feedback with observational data for robust LLM evaluations, and the possibilities for infrastructure that enables this goal.

arXiv

Antares: Foundation Models for Agentic Vulnerability Localization

Supriti Vijay, Aman Priyanshu, Didier Chapoteau, Arthur Goldblatt, Jianliang He, Kimia Majd, Fraser Burch, Baturay Saglam, Takahiro Matsumoto, Zhuoran Yang, amin karbasi

Vulnerability localization is a fundamental step in software security, requiring models to reason over large codebases and iteratively identify vulnerable implementations. We present Antares, a family of compact language models (350M, 1B, and 3B parameters) for agentic vulnerability localization. Based on IBM Granite base models, Antares is trained through a two-stage pipeline that combines supervised fine-tuning on cybersecurity reasoning and repository exploration data with reinforcement learning from verifiable rewards over vulnerable repositories. Across extensive evaluations, Antares-3B approaches GPT-5.5 while outperforming open-weight models over 200x larger in size. The Antares family further enables fast, low-cost local inference, completing a full 500-task evaluation sweep in approximately 15 minutes on a single H100 GPU, corresponding to an amortized evaluation time of under 2 seconds and less than $0.002 per task.

arXiv

Can Foundation Models Hear What Made That Sound? A Tiered Benchmark of Audio-Language Models and Traditional Classifiers for Closed-Set Sound Source Identification

Sajjad Abdoli, Ghassan Al-Sumaidaee, Ahmad ElShiekh, Ahmed Rashad

We benchmark eleven audio classification methods: five task-aware closed-set LLMs (four Gemini models plus open-weight Kimi-Audio-7B-Instruct), four fixed-vocabulary taggers (YAMNet, PANNs, Whisper-AT, and SSLAM), a zero-shot audio-text model (CLAP), and an audio-grounded LLM (BAT). We evaluate them on a closed-set sound-source identification task over 2,242 clips spanning 23 fine-grained classes and 11 categories. Since these methods differ fundamentally in how they receive the task and how outputs are scored, we group them into four evaluation tiers rather than one leaderboard, reporting macro Precision, Recall, F1, and false-negative rate per tier. The best model, Gemini-3.1-Pro-Preview, reaches 85.6 percent category-level F1 and 56.7 percent fine-grained F1. Kimi-Audio is competitive for its size, reaching 67.5 percent category-level F1 and 32.9 percent fine-grained F1, but fails to answer 1.6 percent of samples. SSLAM and CLAP match or exceed the best closed-set model at the category level without seeing the candidate list, but fall behind at the fine-grained level. Analyzing the Gemini models' chain-of-thought across 8,968 responses, we find that response length does not predict accuracy, an apparent "holistic judgment beats detailed analysis" effect is better explained as a difficulty confound, and wrong answers are stated confidently 92 to 100 percent of the time. We report full per-class confusion matrices and metrics for all eleven methods, identify the structural error modes behind most of the accuracy loss between granularities, and give practical guidance for choosing among these method families.

arXiv

Chess on Ice: Curling Tactical Decision-Making via Backward Induction and Deep Reinforcement Learning

Patrick Oberlin, Matteo Cederle, Aren Karapetyan, Saverio Bolognani, Gian Antonio Susto, Florian Dörfler

Curling is often referred to as "Chess on Ice", owing to the tactical complexity of its decision-making process. Yet unlike chess, curling remains largely underexplored from a machine learning perspective, with prior work confined mainly to statistical approaches. We propose a reinforcement learning framework capable of quantitatively evaluating and comparing tactical options in curling. The game poses several modeling challenges: continuous state and action spaces, stochastic action outcomes reflecting player skill variability, and state transitions that are highly sensitive to small perturbations in the executed action. To address them, we employ the Deep Deterministic Policy Gradient actor-critic algorithm, adapted to exploit the finite-horizon structure of the game. Our experiments show that effective curling strategies can be acquired in a fully self-supervised manner, without any human-annotated data: on a reduced four-rock variant, the learned agent matches a hand-crafted expert heuristic in a regime where that heuristic is close to optimal, a parity we quantify against the intrinsic hammer advantage of the variant. Beyond the resulting policy, the learned critic provides a dense value estimate over the entire continuous action space, enabling the quantitative comparison of tactical alternatives for applications such as post-game performance analysis and decision support during athlete preparation.

arXiv

KC-Agent: A Dual-Process Cognitive Architecture for Efficient ML Model Improvement

Gusseppe Bravo-Rocca, Jordi Guitart, Ajay Dholakia, David Ellison, Puneet Jain

Data drift poses significant challenges for machine learning systems in production, requiring continuous model updates to maintain performance. We present KC-Agent, a dual-process cognitive architecture for automated ML model improvement that combines fast pattern recognition (System 1) with deliberate incremental updates (System 2). Our approach implements structured memory systems enabling System 1 to leverage successful solutions previously discovered by System 2, achieving efficient pattern-based responses without costly re-computation. KC-Agent incorporates atomic change principles and rollback capabilities to ensure reliable, verifiable updates in production environments. We evaluate our method on five datasets including real-world NASA turbofan data with authentic temporal degradation and synthetic datasets with controlled drift scenarios. KC-Agent achieves state-of-the-art performance (76.8% accuracy) while maintaining optimal efficiency (13.2s execution time), outperforming established cognitive architectures: CodeAct (+2.4%), Tree of Thoughts (+3.6%), ReAct (+8.0%), and Reflexion (+8.9%). Consensus evaluation by a panel of state-of-the-art LLMs confirms superior strategic efficacy (8.33/10 Smartness score), significantly outperforming baseline agents. The knowledge consolidation mechanism delivers 91% speedup over the slow variant while maintaining higher accuracy. Our approach demonstrates both theoretical foundations and practical viability for cognitive-inspired automated ML improvement systems capable of handling complex real-world data drift scenarios.

arXiv

MechGeo: Autoformalizing and Proving Euclidean Geometry in Lean 4

Hao Shen, Junyu Guo, Tian Cui, Yuxuan Xiao, Lihong Zhi

We present MechGeo, a Mathlib native agentic framework that jointly addresses faithful autoformalization and certified proof construction for Euclidean geometry. In this framework, GeoFormalizer represents informal problems in GeoIR, deterministically translates them into Lean 4, and iteratively repairs candidate statements using structural diagnostics and semantic evaluation. GeoProver constructs geometric proof plans, derives intermediate lemmas, and selectively algebraizes suitable subgoals through a library verified in Lean. Singular or SymPy may generate algebraic certificates, but all resulting proofs and counterexamples are checked by Lean's kernel. Experiments across seven LLM backbones show substantial improvements in autoformalization, particularly for models with weaker direct translation performance. On 43 historical IMO geometry problems, GeoFormalizer generates formal statements that GeoProver proves in 29 cases; for the remaining 14, it constructs counterexamples verified in Lean and proves all repaired statements after expert correction. Together with IMO 2026 Problem 2, this yields, to the best of our knowledge, the largest reported collection of automated, kernel-checked Lean proofs for IMO geometry problems. On the 14 geometry statements in LEAP's Lean-IMO-Bench, MechGeo proves 12 for the first time, formally refutes the remaining two, and proves both repaired statements. These results establish counterexample guided diagnosis, geometric reasoning, and certified symbolic computation as a practical foundation for trustworthy formal geometry.

arXiv

Shared Prefixes, Better Credit: Adaptive Routing for Multi-Agent Reasoning

Yiqing Liu, Zihao Wang, Hantao Yao, Wu Liu, Yongdong Zhang

Multi-agent reasoning (MAR) improves reasoning reliability through iterative solution exchange and refinement. Existing adaptive MAR methods typically learn routing decisions from query-level labels or trajectory-level returns, but such coarse supervision cannot accurately estimate the state-conditioned utility of individual operators in multi-step collaboration. We propose TreeCredit, a shared-prefix credit assignment framework for efficient adaptive MAR. Its core insight is to estimate operator utility through state-matched downstream comparisons, rather than directly attributing trajectory-level outcomes to preceding decisions. TreeCredit constructs shared-prefix collaboration trees by expanding candidate operators from the same intermediate state and assigns each state--operator pair a correctness-prioritized suffix credit based on the terminal correctness and cumulative additional cost of its complete continuation. These structured credits are converted into state-local operator preferences to train a lightweight pairwise state router, which dynamically selects the next admissible operator during inference. Experiments on six reasoning benchmarks show that TreeCredit modestly improves accuracy while substantially reducing inference cost, achieving a better accuracy--cost trade-off than representative MAR methods.

arXiv

SKT: Skill-Use Training at Scale via Verified Synthetic Data Generation

Zelin Tan, Yiqun Zhang, Hao Li, Zhiyao Cui, Hejia Geng, Shao Zhang, Hangfan Zhang, Yang Chen, Xiaosong Wang, Lilong Wang, Zhenfei Yin, Shuyue Hu, Chen Zhang, Lei Bai

Agent skills have become an important mechanism for equipping language-model agents with reusable procedural knowledge. However, providing skills alone does not guarantee that current models can effectively identify, apply, and coordinate them. To improve skill-use capabilities, we introduce SKT, a verified data synthesis pipeline that constructs skill-grounded tasks and executable trajectories from large collections of agent skills. SKT selects suitable single-skill and multi-skill configurations, synthesizes tasks through rule-based and agent-based verification with feedback-guided repair, and retains only successful trajectories that substantially use every required skill. Using 2,000 public skills, SKT produces 4,000 task packages and 27,164 verified trajectories. Based on the same pipeline and a disjoint test pool, we further construct SkillEval, a held-out executable benchmark for evaluating skill use. Experiments across diverse models, benchmarks, and agent harnesses show that supervised fine-tuning on SKT-generated trajectories consistently improves skill-use performance. Verification ablations, cross-harness evaluation, and scaling experiments further demonstrate that these gains depend on high-quality supervision, extend beyond a single agent interface, and increase with broader skill coverage. Together, these results establish verified data synthesis as an effective and scalable approach for skill-use training.

arXiv

Harness-R1: Learning to Edit Executable Runtime Harnesses from Agent Failure Trajectories

Shuai Shao, Kangning Zhang, Qingyao Li, Shijian Wang, Hao Wang, Wenxiang Jiao, Yuan Lu, Yi Guo, Weiwen Liu, Weinan Zhang

Agents built around large language models continually accumulate interaction trajectories during deployment, yet their behavior typically remains fixed. Beyond updating model weights, these trajectories can improve the agent harness that constructs context, mediates tools, validates actions, and recovers execution. We introduce Harness-R1, the first method, to our knowledge, that makes failure-conditioned, lifecycle-wide editing of an existing executable runtime a learned capability. It post-trains a dedicated harness engineer with online reinforcement learning so that its edits are optimized for the realized task success they produce, rather than proposed by a fixed editor. A separate 9B engineer converts batches of target-agent failures into validated executable patches; fresh same-batch reruns of the frozen target provide outcome rewards, so training updates only the engineer. Cold-start supervised fine-tuning initializes this editing policy, which is then trained online with group-relative policy optimization. Across WebShop, ALFWorld, and DBBench, Harness-R1 raises vanilla Qwen3.5-9B success from 44.3% to 53.6% (+9.3 percentage points). After direct target-agent fine-tuning, a target-specific engineer raises the average further from 59.2% to 64.2% (+5.0 points); because these gains hold both before and after fine-tuning the target, Harness-R1 points toward co-evolving the harness engineer and the target agent.

arXiv

TS-MAMP: A Remanufactured Agricultural Robot Powered by Second-Life EV Components and NMS-Free On-Device Weed Detection

Weijie Shi, Zicheng Xu, Zhenbang Cheng, Haoran Xuan, Mingbo Duan, Gan Ge

Agriculture 4.0 robotic systems improve field efficiency yet remain too capital-intensive for the fragmented smallholdings that dominate global agriculture. Meanwhile, a growing number of retired low-speed electric-vehicle (LSEV) powertrains retain functional electromechanical value but are destructively recycled. This paper presents TS-MAMP (Telescopic-Sleeve Modular Agricultural Mobile Platform), a remanufactured robot built under 3R (reduce, reuse, recycle) circular-economy principles. Retired 48 V brushless-DC (BLDC) hub motors are paired via back-EMF matching, and lead-acid battery modules screened at 60%-80% state of health are actively balanced within a 100 mV inter-module voltage deviation. Together, these reused components reduce the powertrain-and-chassis BOM cost by approximately 60%, to below USD 450 (perception and weeding modules excluded). The truss chassis provides >=200 kg static load, continuously adjustable track width from 1200 mm to 2000 mm, and <=5-minute module changeover. An NMS-free (non-maximum-suppression-free) YOLOv10n detector with consistent dual-assignment training and negative-sample learning achieves 80.87% mean average precision (mAP)@0.5 (58.41% mAP@0.5:0.95) on the Wanxi Crop-Weed dataset, and is deployed via FP16 TensorRT on a Jetson Nano, confirming on-device inference feasibility. TS-MAMP demonstrates that retired EV components, under modest screening, can be re-engineered into affordable, AI-enabled agricultural robots--opening a remanufacturing pathway for the smallholder fields that commercial automation leaves unserved.

arXiv

An Autonomous Scientific Knowledge Generation Framework for AI-Driven Scientific Discovery

Dibakar Datta

Artificial intelligence (AI) is transforming scientific discovery, but its effectiveness is fundamentally limited by the availability of structured scientific knowledge. Although existing databases have accelerated data-driven materials research, much of the knowledge needed for predictive modeling and inverse design remains embedded in unstructured scientific literature. We present an Autonomous Scientific Knowledge Generation Framework that transforms scientific publications into a Unified AI-Ready Scientific Knowledge Base. The framework integrates ontology-guided literature acquisition, hybrid scientific knowledge extraction, semantic harmonization, knowledge fusion, and validation within a unified workflow. Rather than treating literature retrieval, information extraction, and database construction as separate tasks, the framework progressively converts scientific publications into structured, semantically consistent, and provenance-preserving knowledge suitable for AI-driven reasoning. As a proof of concept, the framework was applied to electro-optic materials. Autonomous literature acquisition retrieved and validated about 1,000 publications from multiple scholarly repositories. A representative subset of eight publications was processed through the complete workflow, generating 29 structured scientific records that were harmonized into 7 canonical scientific records. The results demonstrate the complete transformation from scientific literature to an AI-ready scientific knowledge base while preserving quantitative measurements, operating conditions, provenance, and scientific context. The proposed framework provides a scalable, domain-independent foundation for predictive AI, generative AI, and closed-loop AI-driven scientific discovery.

arXiv

AgentCompile: An LLM-Guided Compiler for Direct CUDA Inference

Xuanzhe Li, Ziyan Weng, Zhiyu Zhu, Junhui Hou

Transformer inference increasingly relies on specialized compiler and runtime support, while recent LLMs can generate nontrivial CUDA kernels. However, unconstrained generation guarantees neither correctness nor performance. We present \textsc{AgentCompile}, an LLM-guided CUDA inference compiler that combines two complementary uses of LLMs. First, the LLM provides advisory metadata for compiler-derived region summaries and bounded candidate spaces. The compiler then instantiates template-based CUDA candidates, validates correctness, selects implementations by measured latency, and falls back when specialization is unsupported or unprofitable. Second, under compiler-defined contracts, the LLM directly generates five classes of decode-critical kernels to accelerate inference, prompted by distilled optimization principles. \textsc{AgentCompile} integrates these kernels into a serving runtime with paged KV cache, continuous batching, preemption, chunked prefill, and bucketed full-step CUDA Graph replay. Across six evaluated model families, \textsc{AgentCompile} achieves speedups of \textbf{2.23--6.98$\times$} over PyTorch eager for single-request generation, and \textbf{1.04--1.16$\times$} over vLLM for both single-request generation and multi-request serving. Our code is publicly available at https://github.com/veneno1213822/AgentCompile.

arXiv

HUSH-Bench: Measuring Memory-Use Boundaries for Sensitive History in Conversational Agents

Lingxiang Xu, Jiaoyun Yang, Min Hu, Ning An

Long-term memory helps conversational agents maintain continuity across sessions, while relevance and current-turn warrant remain distinct decisions. We study this boundary under a stated conservative policy in which sensitive history shapes a response when the current turn supplies a reason to use it. We introduce HUSH-Bench, a controlled benchmark of 2,400 benign prompts paired with histories containing one marked sensitive disclosure and matched no-memory references. HUSH-Bench measures unsolicited history integration with the Unsolicited History Integration Score (UIS; 0--100, higher is worse), records whether the marked disclosure reaches the generator, and includes paired prompts that differ only in whether the user asks the assistant to use earlier context. We evaluate four models under no-memory, full-context, and three retrieval-based memory settings. Memory access raises UIS from near zero to 8.9--26.6 for one model and 51.3--83.0 for the other three. Retrieval systems expose the marked disclosure in 23.0\%--30.3\% of cases, while related sensitive entries or summaries remain available and three models continue to show high UIS. Across four generators, an explicit invitation increases target-memory uptake scores by 27.0--41.3; measured helpfulness remains stable while mean over-scope rises. These results motivate treating memory storage, retrieval, warrant, and per-turn scope as separate design decisions.

arXiv

Answer Presence Drives RAG Rewriting Gains

Yuejie Li, Yueying Hua, Ke Yang, Li Zhang, Yueping He, Ruiqi Li, Bolin Chen, Tao Wang, Bowen Li, Chengjun Mao

Retrieval-augmented QA pipelines often route retrieved passages through an LLM \emph{rewriter} before a smaller reader, lifting F1 by tens of points on multi-hop benchmarks; this gain is typically credited to improved evidence quality. We ask whether that lift is causally driven by the gold answer string appearing in the rewritten context rather than by curation per se, using a controlled intervention audit. For each rewritten context we re-run the reader after one of four controlled edits to the compile output: removing the gold answer span, replacing a length-matched random non-answer span (placebo), or injecting the gold into rewrites where it was absent (at the prefix or at a midpoint sentence boundary). Across twelve completed (cell, baseline) intervention runs spanning three reader families (Qwen2.5-7B, Qwen3.5-35B, GLM-4.7), two datasets (HotpotQA, 2WikiMultihopQA), and three compiler arrangements (MA-only, MB-only, MA$+$verify), removing the gold answer drops reader F1 by $28$ to $64$ points beyond the length-matched placebo on paired \texttt{answer-in-compile} strata, and prepending the gold into rewrites that lacked it raises F1 by $+0.7$ to $+9.7$ points in $10$ of $12$ (cell, baseline) combinations. A companion five-sentinel audit shows the conventional single-\texttt{[MASK]} probe is itself sentinel-fragile: on 2Wiki it reports a $+4.12$~F1 ``non-leakage residual'' that flips to $-3.33$ to $-7.81$~F1 under four alternative sentinels and fails an equivalence test for three of those four ($1/4$~pass). We do not propose a new rewriter or mitigation; we release the intervention runner and the sentinel panel so that other rewriter-gain claims can be tested against the same standard.

arXiv

Prompt Codebooks: Discrete Compositional Optimization for Language Model Instruction Refinement

Jyotirmoy Nath, Neeraj Kumar, Brejesh Lall

Automatic prompt optimization (APO) has driven significant gains in LLM-based agentic workflows. However, most existing methods treat each task's prompt as a monolithic, instance-blind string optimized through global edits, producing brittle updates and preventing the reuse of learned sub-behaviors. We propose Prompt Codebook Optimization (PCO), a novel compositional prompt optimization framework that recasts APO as discrete learning over a finite vocabulary of natural-language instincts--atomic, reusable instruction units. PCO organizes prompt-construction knowledge in a discrete codebook and routes each input to a small subset of entries via an LLM-based encoder; a generator composes them into a prompt for the executor; a critic emits a structured verdict that decomposes by attribution into per-variable textual gradients, jointly training the encoder, generator, critic, and codebook under a language-valued min-max objective. The resulting routing is per-instance: different inputs in the same task receive different instinct compositions. Across six benchmarks, PCO improves aggregate performance over zero-shot by +13.50 points on Qwen3-8B and +11.80 points on LLaMA-3.1-8B. Crucially, PCO surpasses GEPA on HotpotQA by +6.34 points (Qwen3-8B) and +4.27 points (LLaMA-3.1-8B), while simultaneously reducing deployed prompt length by up to 14.1$\times$ vs. MIPROv2 and 3.0$\times$ vs. GEPA.

arXiv

AST: Adaptive, Seamless, and Training-Free Precise Speech Editing

Sihan Lv, Yechen Jin, Zhen Li, Jintao Chen, Jinshan Zhang, Ying Li, Jianwei Yin, Meng Xi

Text-based speech editing aims to modify specific segments while preserving speaker identity and acoustic context. Current approaches generally involve either expensive task-specific training or adapting pre-trained Text-to-Speech (TTS) models. However, both paradigms face challenges: task-specific methods often degrade fidelity in unedited regions, whereas TTS adaptations struggle with a trade-off between editing naturalness and temporal fidelity. To address these issues, we propose AST, an Adaptive, Seamless, and Training-free speech editing framework. Built upon pre-trained TTS, AST leverages Latent Recomposition to stitch preserved source segments with synthesized targets, guaranteeing fidelity in unedited regions. To break the quality-controllability trade-off, we introduce Adaptive Weak Fact Guidance (AWFG), which modulates a mel-space signal to ensure seamless boundary transitions without disrupting the generative manifold. Furthermore, to address evaluation gaps in temporal fidelity, we propose a new benchmark suite: the LibriSpeech-Edit dataset and a novel metric, Word-level Dynamic Time Warping (WDTW). Extensive experiments demonstrate that AST consistently outperforms existing task-specific and fine-tuned speech editing methods across content accuracy, perceptual quality, speaker preservation, and temporal fidelity. Remarkably, AST achieves state-of-the-art zero-shot speech editing performance without any task-specific training or paired editing data, validating the effectiveness of latent recomposition and AWFG in bridging the quality-controllability trade-off.

arXiv

Group Selection as a Safeguard Against AI Substitution

Qiankun Zhong, Thomas F. Eisenmann, Julian Garcia, Iyad Rahwan

Reliance on generative AI can reduce cultural variance and diversity, especially in creative work. This reduction in variance has already led to problems in model performance, including model collapse and hallucination. In this paper, we examine the long-term consequences of AI use for human cultural evolution and the conditions under which widespread AI use may lead to "cultural collapse", a process in which reliance on AI-generated content reduces human variation and innovation and slows cumulative cultural evolution. Using an agent-based model and evolutionary game theory, we compare two types of AI use: complement and substitute. AI-complement users seek suggestions and guidance while remaining the main producers of the final output, whereas AI-substitute users provide minimal input, and rely on AI to produce most of the output. We then study how these use strategies compete and spread under evolutionary dynamics. We find that AI-substitute users prevail under individual-level selection despite the stronger reduction in cultural variance. By contrast, AI-complement users can benefit their groups by maintaining the variance needed for exploration, and can therefore be favored under cultural group selection when group boundaries are strong. Overall, our findings shed light on the long-term, population-level effects of AI adoption and inform policy and organizational strategies to mitigate these risks.

arXiv

Scaling an Autoregressive Transformer for Single-Cell Generation

Aleksandr Sharipov, Yusif Mukhtarov, Igor Molybog

We study a self-supervised generation task for single-cell gene expression vectors: given a set of vectors from a cell type, we aim to generate additional gene expression vectors of that cell type. For this task we characterize both the biological fidelity of the generated gene expression vectors and the scaling behavior of the pretraining loss. The model is a causal transformer paired with a learned quantized VAE tokenizer, trained with a cross-entropy loss. To evaluate the model, we condition it on held-out gene expression vectors of a cell type and generate vectors of gene expression, comparing the resulting distribution over gene expression vectors to the ground truth distribution of that cell type. We study the scaling properties of the proposed architecture by varying the number of trained parameters and the amount of training data. To our knowledge, we find the first jointly-fit two-exponent scaling law and compute-optimal frontier for a single-cell foundation model. Finally, we discuss how this pretrained model could be finetuned for perturbation response prediction.

arXiv

Inverted Detection and Control in Steering Vectors

Max Torop, Aria Masoomi, Jennifer Dy

Steering vectors (SVs) are widely used to influence the expression of concepts (e.g., truthfulness) in large language model outputs. A key assumption underpinning SVs is that they are linearly discriminative with respect to the concept: representations of texts that exhibit the concept are more aligned with the SV than those that do not, motivating shifts along the positive or negative SV direction to respectively promote or suppress the concept. In this work, we identify an inverted detection-control phenomenon in which some highly discriminative SVs that are aligned with positive representations can consistently promote the opposite behavior. We refer to such vectors as inverted-steering vectors (ISVs). We provide a geometric characterization of ISVs' effects, finding that steering along these directions systematically pushes representations in discriminative downstream heads as if the concept were absent, even prior to decoding. Motivated by this analysis, we propose an approach for distinguishing ISVs without requiring generation or associated response scoring. This enables targeted sign flips, which we use to improve a foundational detection-based steering pipeline via Inference Time Intervention (ITI). Our approach improves results in 27/30 experiments, ranging from +0.9% to +138%. We evaluate our findings on Gemma 3 12B, Qwen 2.5 14B, and Olmo 3 7B across 5 concepts.

arXiv

SP3O: Reinforcement Learning from Segment Preferences without Reward Modeling

Evan Assmus, Qining Zhang, Lei Ying

Preference-based reinforcement learning (PbRL) for general stochastic MDPs often requires training a reward model. Existing reward-model-free methods are either restricted to bandits or deterministic MDPs, such as DPO or P3O, or use zeroth-order, gradient-free optimization, which in general exhibits a slower convergence rate than gradient-based algorithms. Furthermore, existing reward-model-free preference-based RL algorithms almost exclusively use trajectory-level feedback, which can require significant effort from a human evaluator when trajectories are long. On the other hand, segments are much shorter, so they are easier to compare and evaluate. In this paper, we introduce a novel reward-model-free, critic-free, and gradient-based PbRL algorithm compatible with segment preferences named Segment Pairwise Proximal Policy Optimization (SP3O). SP3O utilizes segment-level preference feedback to construct an accurate policy value difference estimator via off-policy importance sampling, and then uses the estimator to compute the policy gradient via a PPO-type loss function. We provide a theoretical basis for the algorithm and analyze the tradeoff in choosing the segment length. We also evaluate it experimentally against other PbRL/RLHF algorithms in robotic control and LLM finetuning settings to show its improved performance, especially in long-horizon tasks.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.