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.

ICML 2026

SE(n)-Invariant Flow Matching: A General Framework with Application to Object Reassembly

Gaël Heck, Sylvie Le Hégarat-Mascle, Nicolas Lermé

Reassembling $N$ fragments in $n$-dimensional space is a shape reconstruction task that is invariant to global rigid motions. Training directly on $\mathcal{M}=\mathrm{SE}(n)^N$ can be ill-posed: standard losses penalize solutions that differ only by a global transform. Existing methods often address this with ad-hoc anchoring which breaks permutation invariance across fragments and can introduce biases that must be mitigated with extensive and costly data augmentation. We propose a geometric framework that enforces invariance by construction. First, a **Global Gauge Fixing** (GGF) strategy deterministically aligns configurations using an intrinsic generalized-inertia rule. Second, we introduce a **quotient-invariant Flow Matching objective** that operates via orthogonal projection onto the horizontal tangent bundle. This construction factors out global pose at each timestep, enabling the model to learn only shape-changing dynamics on the quotient space $\mathcal{M}/\mathrm{SE}(n)$. Our unified $\mathrm{SE}(n)$-invariant framework admits efficient closed-form 2D/3D instantiations and improves accuracy on polygonal jigsaw puzzles and 3D fracture reassembly benchmarks.

ICML 2026

HELIX: Hybrid Encoding with Learnable Identity and Cross-dimensional Synthesis for Time Series Imputation

Fengming Zhang, Wenjie Du, Huan Zhang, Ke Yu, Shen Qu

Time series imputation benefits from leveraging cross-feature correlations, yet existing attention-based methods re-discover feature relationships at each layer, lacking persistent anchors to maintain consistent representations. To address this, we propose HELIX, which assigns each feature a learnable feature identity, a persistent embedding that captures intrinsic semantic properties throughout the network. Unlike graph-based methods that rely on predefined topology and assume homogeneous spatial relationships, HELIX learns arbitrary feature dependencies end-to-end from temporal co-variation, naturally handling datasets where features mix spatial locations with semantic variables. Integrated with hybrid temporal-feature attention, HELIX achieves the state-of-the-art performance, surpassing all 16 baselines on 5 public datasets across 21 experimental settings in our evaluation. Furthermore, our mechanistic analysis reveals that HELIX aligns learned feature identities and dependencies with latent physical and semantic structure progressively across layers, demonstrating that it more effectively translates cross-feature structure into imputation accuracy.

ICML 2026

Manifold-Aligned Guided Integrated Gradients for Reliable Feature Attribution

Soyeon Kim, Seongwoo Lim, Kyowoon Lee, Jaesik Choi

Feature attribution is central to diagnosing and trusting deep neural networks, and Integrated Gradients (IG) is widely used due to its axiomatic properties. However, IG can yield unreliable explanations when the integration path between a baseline and the input passes through regions with noisy gradients. While Guided Integrated Gradients reduces this sensitivity by adaptively updating low-gradient-magnitude features, input-space guidance still produces intermediate inputs that deviate from the data manifold. To address this limitation, we propose **Manifold-Aligned Guided Integrated Gradients** (MA-GIG), which constructs attribution paths in the latent space of a pre-trained variational autoencoder. By decoding intermediate latent states, MA-GIG biases the path toward the learned generative manifold and reduces exposure to implausible input-space regions. Through qualitative and quantitative evaluations, we demonstrate that MA-GIG produces faithful explanations by aggregating gradients on path features proximal to the input. Consequently, our method reduces off-manifold noise and outperforms prior path-based attribution methods across multiple datasets and classifiers. Our code is available at https://github.com/leekwoon/ma-gig/.

ICML 2026

OC-space: a Unifying Perspective on Verification of Tree Ensembles

Timo Martens, Laurens Devos, Lorenzo Cascioli, Wannes Meert, Hendrik Blockeel, Jesse Davis

We study the problem of verifying whether certain properties such as robustness or fairness hold in an ensemble of decision trees. This problem is known to be NP-hard, with most research targeting a solution to a specific verification task. We explore the problem through the lens of an ensemble's OC-space: the set of all possible combinations of the individual trees' predictions. This provides a unifying view that yields a more generic and flexible approach to verification. We show that a wide variety of existing verification tasks can be (1) framed as simple searches through OC-space, and (2) answered in time linear or quadratic in the size of the OC-space. Moreover, the search can be made more efficient by using spatial index structures. Interestingly, while the OC-space can grow exponentially with the ensemble's size, in practice it is often feasible to enumerate all output configurations. Empirically, we show that our generic approach can be faster than approaches targeting a single verification task.

ICML 2026

Steering Beyond the Support: Adversarial Training on Unsupervised Jailbroken Activation Simulation

LUOYU CHEN, Weiqi Wang, Zhiyi Tian, Chenhan Zhang, Feng Wu, Jianhuan Huang, Ahmed Asiri, Shui Yu

Jailbreak prompts can trigger harmful completions on aligned LLMs, In accordance, safety steering has been proposed: test-time activation interventions that steer jailbreak activations to trigger refusal while preserving benign utility. However, existing steering methods are fundamentally supervised and tied to a static, limited training set, whereas real jailbreaks evolve and are often out-of-distributed from the training set, leading to failures on unseen attacks. In this paper, we tackle this failure by developping a \emph{zero-shot} defense. Base on unsupervised latent direction discovery, we directly simulate jailbroken activations without any knowledge of jailbreak strategy. To build a defense mechnism upon this, we propose a bi-level adversarial training framework. In the inner step, we simulate diverse jailbroken activations by extrapolating from refusal state harmful-request activations via unsupervised latent direction discovery. In the outer step, we train a potential-induced steering field to push these adversarial jailbroken states into refusal regions while keeping benign unchanged. Across three LLMs and six classical jailbreak families, our method achieves strong defense with attack success rates mostly below 5%, and we analyzed the increasing subspace coverage of our simulated jailbroken activations on real jailbreaks throughout training, which helps explain the increasing robustness of our defense mechnism.

ICML 2026

3D MeanFlow: One-Step Point Cloud Completion and Generation via Average-Velocity Transport

Haowen Zhong, Jiujun Cheng, Haowen Wang, Chao Wei, lu Yang, Qichao Mao, Shangce Gao

Point cloud completion and generation are important across many 3D tasks, where both fidelity and sampling efficiency matter. Prevailing high-fidelity approaches rely on long sampling schedules, which incur substantial inference latency. Few-step alternatives typically use rectification or distillation, leading to multi-stage training pipelines and potential quality trade-offs. We present 3D MeanFlow (3DMF), a distillation-free model that performs one-step average-velocity transport for point cloud completion and generation. We optimize an instantaneous-average consistency objective and impose a shape-level constraint to stabilize training. Additionally, we introduce PointPlug, integrating completion into 3D object detectors and evaluating its impact. PointPlug uses adaptive selection that balances benefit and latency. Across standard benchmarks, 3DMF achieves one-step sampling with an order-of-magnitude speedup while maintaining competitive fidelity. On nuScenes and KITTI, inserting PointPlug improves all evaluated detectors under comparable settings.

ICML 2026

EquiCAD: A Geometric Equivariant Neural Network for 3D Shape Classification

Yonghao Su, Yantao Gan, Junfeng Long, Caiyang Yu, Wenhao Zheng, Jiancheng Lv, Xianggen Liu

Three-dimensional (3D) shape classification plays a central role in computer vision and computer-aided design (CAD), underpinning applications in intelligent manufacturing, automated inspection, and digital engineering. Despite recent progress with 3D CNNs and graph-based approaches, existing methods often overlook the geometric-topological regularities and symmetry principles intrinsic to CAD boundary representations (B-reps). To address this challenge, we introduce EquiCAD, a symmetry-aware learning framework that integrates equivariant representations with graph-based reasoning. By leveraging group-theoretic decomposition of curve and surface descriptors, EquiCAD preserves symmetry-consistent feature transformations while retaining rich geometric details. The model further exploits hierarchical message passing to capture interactions between local features and global structure. Experimental results across multiple datasets, including SolidLetters, Parts, the Machining Feature benchmark, and our newly constructed Features dataset, demonstrate substantial improvements over prior state-of-the-art approaches, particularly on industrially relevant shapes with fine-grained attributes. These findings highlight the value of symmetry-aware modeling for robust and generalizable 3D shape analysis.

ICML 2026

Universal Learning of Nonlinear Dynamics

Evan Dogariu, Anand Paresh Brahmbhatt, Elad Hazan

We study the fundamental problem of one-step prediction of a marginally stable unknown nonlinear dynamical system. We describe an algorithm for this problem, based on the technique of spectral filtering, which learns a mapping from past observations to the next based on a spectral representation of the system. Using techniques from online convex optimization, we prove vanishing prediction error for any nonexpansive nonlinear dynamical system with finitely many marginally stable modes, with rates governed by a novel quantitative control-theoretic notion of learnability. The main technical component of our method is a new spectral filtering algorithm for linear dynamical systems, which incorporates past observations and applies to general noisy and marginally stable systems. This generalizes the original spectral filtering algorithm to both asymmetric dynamics as well as incorporating noise correction, and is of independent interest.

ICML 2026

Beyond Structural Symmetries: Linear Mode Connectivity via Neuron Identifiability

Vincent Bürgin, Daniel Herbst, Ya-Wei Eileen Lin, Ya-Wei Eileen Lin, Stefanie Jegelka

Many striking phenomena in deep learning, such as linear mode connectivity and the structured behavior of training dynamics, are closely tied to parameter symmetries: transformations that leave the realized function unchanged. Despite growing attention to parameter symmetries, the exact interplay between parameters, data, and representations remains underexplored. To investigate this, we develop a theoretical framework of effective function classes, i.e., the set of functions a neuron can realize on its input support, and the norm cost of realizing them. We then formalize *effective symmetry breaking* via neuron identifiability across independent training runs. Our analysis shows that neural networks can admit large families of approximately equivalent solutions even in *structurally asymmetric* models. We further show that neuron identifiability enables representation merging *without prior alignment*, and characterize when such merging admits a linear low-loss path. These findings highlight the role of effective function classes in affecting the loss landscape.

ICML 2026

StructMAR: Structure-Aware Masked Autoregression for Explicit Layout Alignment in Text-to-Image Generation

Gang Cao, Junying Zhang

Although text-to-image generation has achieved significant progress, strict instance-level layout alignment remains a challenge for many applications. Masked Autoregressive (MAR) models on continuous latents are both efficient and high-fidelity, yet the standard practice of flattening 2D latents into 1D sequences weakens spatial topology, limiting precise controllability. To address this, we propose StructMAR, a structure-aware masked autoregressive framework that transforms layout alignment from a soft correlation into an explicit structural alignment. By integrating 2D Rotary Positional Embeddings with a Layout-Guided Attention Bias, StructMAR explicitly biases latent tokens toward their corresponding layout instances during attention computation. We further use Group Relative Policy Optimization (GRPO) as a final-stage policy refinement to reduce the mismatch between the MAR training objective and detector-based layout evaluation metrics. Evaluated on the COCO-Position and COCO-MIG benchmarks, StructMAR achieves state-of-the-art performance, reaching 57.2 AP and 79.4 mIoU on the former, and 61.7 ISR and 56.9 mIoU on the latter. These results, coupled with a 4.05$\times$ inference speedup, underscore the efficacy of explicit structural inductive biases in controllable autoregressive generation.

ICML 2026

Hard Labels In! Rethinking the Role of Hard Labels in Mitigating Local Semantic Drift

Jiacheng Cui, Bingkui Tong, Xinyue Bi, Xiaohan Zhao, Jiacheng Liu, Zhiqiang Shen

Soft labels from teacher models are a de facto practice for knowledge transfer and large-scale dataset distillation (e.g., SRe$^2$L, LPLD). However, when we limit the number of crops per image to reduce the substantial cost of storing precomputed soft labels, these methods suffer severely from local semantic drift: visually ambiguous crops can cause soft supervision to deviate from the image-level ground-truth semantics, leading to persistent errors and a train–test distribution mismatch. We revisit the overlooked role of hard labels and show that, when properly integrated, they can act as a content-invariant semantic anchor that calibrates such drift. We theoretically analyze the emergence of drift under sparse soft-label supervision and demonstrate that hybridizing hard and soft labels restores alignment between visual content and semantic supervision. Building on this insight, we propose a new training paradigm, Hard Label for Alleviating Local Semantic Drift (HALD), which uses hard labels as intermediate corrective signals while preserving the fine-grained benefits of soft labels. Extensive experiments on dataset distillation and large-scale classification benchmarks show consistent generalization improvements. On ImageNet-1K, our method achieves 42.7% accuracy with only 285M soft-label storage (reduces by 100$\times$), outperforming prior state-of-the-art LPLD by 9.0%. Code is available at https://github.com/Jiacheng8/HALD.

ICML 2026

Bridging Your Imagination with Audio-Video Generation via a Unified Director

Jiaxu Zhang, Tianshu Hu, Yuan Zhang, Zenan Li, Linjie Luo, Mingyuan Gao, Guosheng Lin, Xin Chen

Existing AI-driven video creation systems typically treat script drafting and key-shot design as two disjoint tasks: the former relies on large language models, while the latter depends on image generation models. We argue that these two tasks should be unified within a single framework, as logical reasoning and imaginative thinking are both fundamental qualities of a film director. In this work, we propose **UniMAGE**, a unified director model that bridges user prompts with well-structured scripts, thereby empowering non-experts to produce long-context, multi-shot films by leveraging existing audio–video generation models. To achieve this, we employ the Mixture-of-Transformers architecture that unifies text and image generation. To further enhance narrative logic and keyframe consistency, we introduce a ``first interleaving, then disentangling" training paradigm. Specifically, we first perform **Interleaved Concept Learning**, which utilizes interleaved text–image data to foster the model’s deeper understanding and imaginative interpretation of scripts. We then conduct **Disentangled Expert Learning**, which decouples script writing from keyframe generation, enabling greater flexibility and creativity in storytelling. Extensive experiments demonstrate that UniMAGE achieves state-of-the-art performance among open-source models, generating logically coherent scripts and visually consistent keyframe images.

ICML 2026

Preserve-Then-Quantize: Balancing Rank Budgets for Quantization Error Reconstruction in LLMs

Yoonjun Cho, Dongjae Jeon, Soeun Kim, Moongyu Jeon, Albert No

Quantization Error Reconstruction (QER) reduces accuracy loss in Post-Training Quantization (PTQ) by approximating weights as $\mathbf{W} \approx \mathbf{Q} + \mathbf{L}\mathbf{R}$, using a rank-$r$ correction to reconstruct quantization error. Prior methods devote the full rank budget to error reconstruction, which is suboptimal when $\mathbf{W}$ has intrinsic low-rank structure and quantization corrupts dominant directions. We propose Structured Residual Reconstruction (SRR), a rank-allocation framework that preserves the top-$k$ singular subspace of the activation-scaled weight before quantization, quantizes only the residual, and uses the remaining rank $r-k$ for error reconstruction. We derive a theory-guided criterion for selecting $k$ by balancing quantization-exposed energy and unrecoverable error under rank constraints. We further show that resulting $\mathbf{Q}+\mathbf{L}\mathbf{R}$ parameterization naturally supports Quantized Parameter-Efficient Fine-Tuning (QPEFT), and stabilizes fine-tuning via gradient scaling along preserved directions. Experiments demonstrate consistent perplexity reductions across diverse models and quantization settings in PTQ, along with a 5.9 percentage-point average gain on GLUE under 2-bit QPEFT. The project page is available at https://ai-isl.github.io/srr.

ICML 2026

OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text- and Time-Series Data

Patrick Langer, Thomas Kaar, Max Rosenblattl, Maxwell A. Xu, Winnie Chow, Martin Maritsch, Robert Jakob, Ning Wang, Juncheng Liu, Aradhana Verma, Brian Han, Daniel Seung Kim, Henry Chubb, Scott R. Ceresnak, Aydin Zahedivash, Alexander T Sandhu, Fatima Rodriguez, Daniel McDuff, Elgar Fleisch, Oliver Oppers Aalami, Filipe Barata, Paul Schmiedmayer

Large Language Models have shown strong capabilities in interpreting multimodal data but remain limited in handling time-series data. Addressing this gap could help to translate longitudinal and wearable data into actionable insights and patient-facing digital health applications. We propose OpenTSLM, an open-source family of Time Series Language Models integrating time-series as a native modality into pretrained LLMs, enabling natural-language prompting and reasoning over multiple time-series via either soft prompting (OpenTSLM-SoftPrompt) or cross-attention (OpenTSLM-Flamingo). To enable training models for time-series reasoning, we introduce three datasets: HAR-CoT (human activity recognition), Sleep-CoT (sleep staging), and ECG-QA-CoT (electrocardiogram question answering). Across tasks, OpenTSLM models outperform baselines, reaching F1 scores of 69.88% in sleep staging and 67.64% in HAR; OpenTSLM-Flamingo also scales more efficiently in memory as the number and length of time series increase. Expert evaluations with cardiologists show that OpenTSLMs exhibit strong reasoning capabilities on raw ECG data.

ICML 2026

From Optimization to Generalization under Heavy-Tailed Data: The Role of Gradient Clipping

Aleksandr Shestakov, Martin Takáč, Eduard Gorbunov

Gradient clipping is widely used to stabilize stochastic gradient methods and is often theoretically motivated by heavy-tailed gradient noise, where even second moments may be infinite, seemingly contradicting the finite-sum ERM setting, where all empirical moments are finite once the dataset is fixed. We resolve this paradox by explicitly separating data sampling from optimization randomness: although moments are finite conditional on the dataset, heavy-tailed data induce dataset-dependent noise whose second moment typically grows with the dataset size $N$. In particular, when $\|\nabla f(x_\star,\xi)\|$ has tail index $\alpha \in (1,2)$, the quantity $\frac{1}{N}\sum_{i=1}^N\|\nabla f(x_\star,\xi_i)\|^2$ scales as $N^{\frac{2}{\alpha}-1}$, leading to deteriorating convergence guarantees for standard SGD as $N$ increases. In contrast, we show that SGD with clipping avoids this growth and admits finite-sum convergence guarantees under heavy-tailed data for broad step-size and clipping schedules. We further derive generalization bounds for strongly convex smooth objectives and show that the tail behavior of gradients at the population minimizer is the key quantity linking optimization and generalization under heavy-tailed data.

ICML 2026

When Sample Selection Bias Precipitates Model Collapse

Xinbao Qiao, Xianglong Du, Wei Liu, Jingqi Zhang, Peihua Mai, Meng Zhang, Yan Pang

The proliferation of recursive training on synthetic data can alleviate data scarcity but risks model collapse, where repeated training erodes distributional tails and homogenizes outputs. Data selection is widely viewed as a remedy, yet its reliability depends critically on the reference distribution used by the verifier. We show that in low-resource verification regimes, where each verifier observes only a small, fragmented, and biased slice of the target manifold, selection itself becomes biased. This situation naturally arises in low-resource data silos such as healthcare consortia or proprietary financial institutions, where raw data cannot be pooled and local references are inherently incomplete. As a result, selection preferentially retains samples aligned with the local manifold while pruning globally relevant tail modes, turning from a safeguard against collapse into a mechanism that precipitates it. We theoretically prove that such siloed selection accelerates collapse and induces power-law diversity decay. As an initial mitigation, we construct Wasserstein proxy references from multiple silos without sharing raw data. Empirical results confirm that local-reference selection fails on skewed distributions, whereas collaborative proxy references mitigate diversity degradation, suggesting that recursive synthetic-data pipelines require particular caution when real-data coverage is fragmented or scarce.

ICML 2026

Distinguishable Deletion: Unifying Knowledge Erasure and Refusal for Large Language Model Unlearning

Puning Yang, Junchi Yu, Qizhou Wang, Philip Torr, Bo Han, Xiuying Chen

Mitigating sensitive and harmful outputs is fundamental to ensuring safe deployment of LLMs. Existing approaches typically follow two paradigms: Knowledge Deletion (KD), which erases undesirable information during training, and Distinguishable Refusal (DR), which steers models away from using sensitive knowledge during inference. Despite rapid progress, KD-based unlearning struggles with biased deletion due to suppressing specific token sequences as a substitute for complete knowledge removal, whereas DR-based unlearning risks the re-emergence of harmful knowledge because the underlying knowledge remains intact. To address these issues, we propose Distinguishable Deletion ($\mathrm{D^2}$), a paradigm that restricts the response distribution in the latent representation rather than specific tokens to erase undesirable knowledge, while distinguishing it from retained knowledge, enabling a refusal mechanism to handle unlearned inputs safely and coherently. To implement $\mathrm{D^2}$, we introduce an energy index that quantifies the presence of knowledge and the separation between unlearned and retained content. Mathematical and empirical analyses show that energy is both accurate and efficient, enabling Energy-based Unlearning Alignment (EUA) to enforce energy-boundary unlearning during training and apply an energy-based refusal mechanism at inference. Extensive experiments demonstrate that EUA significantly outperforms previous methods, indicating the superiority of $\mathrm{D^2}$. Our code is available at https://github.com/Puning97/EUA-for-LLM-Unlearning.

ICML 2026

Decomposing the Basic Abilities of Large Language Models: Mitigating Cross-Task Interference in Multi-Task Instruct-Tuning

Bing Wang, Ximing Li, Changchun Li, Jinjin Chi, Gang Niu, Masashi Sugiyama

Recently, the prominent performance of large language models (LLMs) has been largely driven by multi-task instruct-tuning. Unfortunately, this training paradigm suffers from a key issue, named cross-task interference, due to conflicting gradients over shared parameters among different tasks. Some previous methods mitigate this issue by isolating task-specific parameters, e.g., task-specific neuron selection and mixture-of-experts. In this paper, we empirically reveal that the cross-task interference still exists for the existing solutions because of many parameters also shared by different tasks, and accordingly, we propose a novel solution, namely Basic Abilities Decomposition for multi-task Instruct-Tuning (BADIT). Specifically, we empirically find that certain parameters are consistently co-activated, and that co-activated parameters naturally organize into base groups. This motivates us to analogize that LLMs encode several orthogonal basic abilities, and that any task can be represented as a linear combination of these abilities. Accordingly, we propose BADIT that decomposes LLM parameters into orthogonal high-singular-value LoRA experts representing basic abilities, and dynamically enforces their orthogonality during training via spherical clustering of rank-1 components. We conduct extensive experiments on the SuperNI benchmark with 6 LLMs, and empirical results demonstrate that BADIT can outperform SOTA methods and mitigate the degree of cross-task interference.

ICML 2026

Beyond Mode Collapse: Distribution Matching for Diverse Reasoning

Xiaozhe Li, Yang Li, Xinyu Fang, Shengyuan Ding, Peiji Li, Yongkang Chen, Yichuan Ma, TianYi Lyu, Linyang Li, Dahua Lin, Qipeng Guo, Qingwen Liu, Kai Chen

On-policy reinforcement learning methods like GRPO suffer from \emph{mode collapse}: they exhibit reduced solution diversity, concentrating probability mass on a single solution once discovered and ceasing exploration of alternative strategies. We show this stems from reverse KL minimization's mode-seeking behavior, which reinforces the first high-reward trajectory found rather than maintaining a distribution over multiple diverse solutions. We propose DMPO (\textbf{D}istribution-\textbf{M}atching \textbf{P}olicy \textbf{O}ptimization), which prevents mode collapse through principled approximation of forward KL minimization. DMPO constructs a group-level target distribution over sampled trajectories proportional to their rewards, then aligns the policy distribution to this target. This provides mode-covering behavior without requiring sampling from the intractable global target distribution, enabling sustained exploration throughout training. We validate DMPO on NP-hard combinatorial optimization, where exponentially many feasible solutions exist but only a few approach optimality—an ideal testbed for evaluating exploration. DMPO achieves {43.9\% Quality Ratio on text-based NP-Bench (vs. GRPO's 40.1\%)} and {43.1\% on vision-based NP-Bench (vs. 38.4\%)}—demonstrating 9\% and 12\% relative improvements respectively. These gains generalize to mathematical reasoning (+2.0\%) and out-of-domain tasks (+2.3\%), showing that diversity-preserving training enhances general reasoning capabilities across modalities. Our work establishes distribution matching as a practical, principled approach to preventing mode collapse in on-policy RL, with consistent quality improvements demonstrating sustained exploration across diverse reasoning tasks.

ICML 2026

Success-Conditioning as Policy Improvement: The Optimization Problem Solved by Imitating Success

Daniel Russo

A widely used technique for improving policies is success conditioning, in which one collects trajectories, identifies those that achieve a desired outcome, and updates the policy to imitate the actions taken along successful trajectories. This principle appears under many names---rejection sampling with SFT, goal-conditioned RL, Decision Transformers---yet what optimization problem it solves, if any, has remained unclear. We prove that success conditioning exactly solves a trust-region optimization problem, maximizing policy improvement subject to a $\chi^2$ divergence constraint whose radius is determined automatically by the data. This yields an identity: relative policy improvement, the magnitude of policy change, and a quantity we call action-influence---measuring how random variation in action choices affects success rates---are exactly equal at every state. Success conditioning thus emerges as a conservative improvement operator. Exact success conditioning cannot degrade performance or induce dangerous distribution shift, but when it fails, it does so observably, by hardly changing the policy at all. We apply our theory to the common practice of return thresholding, showing this can amplify improvement, but at the cost of potential misalignment with the true objective.

Keep browsing

More papers from top AI conferences

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