NPCI AI Research Team, Aman Kumar, Asit Desai +15cs.AI cs.CL
Banks need conversational systems that can answer product questions, assist customers with account-related requests, and operate safely within strict operational and regulatory constraints. General-purpose language models do not reliably meet these requirements. They fall short when a task requires grounded information, correct tool use, or cautious handling of bank-specific sensitive situations. We introduce FiMI Banking, a controlled Indian retail-banking setting. We build it from vetted banking documents, structured ground truth, synthetic customer backgrounds, and banking tools. We evaluate two post-training approaches: preference optimization for response-level behavior, and reinforcement learning with verifiable rewards for multi-turn tool-use tasks. Preference optimization improves safe behavior substantially: out-of-scope refusal rises from 52% to 80%. Reinforcement learning improves edge-case performance from 0.509 to 0.718 and order-sensitive task performance from 0.590 to 0.679, while using 29% fewer generated tokens. These results show that preference optimization and verifiable-reward reinforcement learning address complementary requirements for reliable banking agents.
Pairwise preference labels rank complete images, yet Diffusion-DPO applies their effect over many spatial and denoising-time coordinates. For attention-based, noise-prediction latent diffusion, ToPO (Token-Oriented Preference Optimization) constructs a per-minibatch, detached, separable spatial-temporal route from branchwise squared-residual contrast in a frozen reference denoiser. Preferred-branch cross-attention uses content tokens to modulate the spatial factor, and an auxiliary pixel-midpoint ordering term is added without local labels or a learned reward model. In matched three-seed retrainings with a shared update schedule, ToPO has higher endpoint estimates than Diffusion-DPO on all five reported SD-1.5 metrics and on HPSv2, ImageReward, and CLIP for SDXL. It also receives larger raw win shares in an aggregate blind SDXL A/B study. These findings are scoped to the reported equal-update U-Net protocols rather than an equal-compute comparison.
People increasingly turn to large language models (LLMs) for everyday advice, making ethically charged interpersonal problems a practical moral-advisory context. Most prior work has studied this context through single-turn judgments or pressure-laden rebuttals, assumptions that poorly match how guidance is sought in real-world contexts. These assumptions leave unclear whether narration alone, without an explicit opposing position, can shift model judgments during multi-turn moral consultation. Yet real-world moral-conflict conversation often elicits one party's self-justifying account, which can unfold over multiple turns and create information asymmetry. We introduce \textbf{narrative captivity}, a failure mode in which a model treats an unopposed one-sided account as complete and aligns with the narrator's interpretation without seeking missing perspectives. To measure this phenomenon, we build a benchmark of $5{,}078$ interpersonal-conflict scenarios spanning six moral dimensions. Across 17 LLMs, narrative captivity is widespread: end-state judgments under multi-turn narration shift by 25 percentage points on average beyond the matched single-turn baseline. Stage-level analysis identifies preference optimization as a major contributor, while four inference-time strategies provide only partial mitigation. We hope our project fosters LLM advisors that preserve independent judgment in real-world consultation.
Model merging provides an efficient paradigm for constructing multi-task large language models (LLMs) without full model retraining, yet it remains challenged by parameter interference. While existing methods aim to preserve the capabilities of individual expert models and mitigate interference, they generally do not directly learn from the potentially degraded behaviors exposed by naive merging. In this paper, we propose a conflict-driven preference optimization framework for model merging (CoMerge), which reformulates model merging as a preference optimization problem. The approach utilizes a self-supervised, conflict-driven strategy that leverages the defects of naive merging methods (e.g., task arithmetic) as hard negative samples to construct preference pairs without external annotations. By applying preference optimization to refine lightweight, tensor-wise merging coefficients, CoMerge enables the model to mitigate parameter-space conflicts while preserving task-specific capabilities. Extensive experiments show that CoMerge achieves an average normalized performance of 0.9968 on MergeBench, outperforming all evaluated data-free and data-driven model-merging baselines. Furthermore, on Llama-3.1-8B-Instruct, CoMerge yields marked improvements on conflict-sensitive tasks such as instruction following and safety, while remaining highly competitive with full-parameter fine-tuning despite optimizing only 1,445 scalar coefficients.
Reliable disaster damage assessment requires models that provide both accurate predictions and transparent explanations. However, existing multimodal approaches are limited by scarce annotated data and insufficient evaluation of reasoning quality. This study proposes a two-stage training framework that integrates Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) within a unified data construction pipeline. From a single Human-in-the-Loop (HITL) annotation workflow, two complementary datasets are derived, namely ReasoningSet, which contains validated rationales for SFT, and PreferenceSet, which comprises paired rationales for DPO-based alignment. The framework evaluates both classification performance and explanation quality using automatic metrics, model-based scoring, and human ranking. Experimental results show that SFT improves accuracy from 73.64% to 78.29% and increases Macro-F1 by 29% compared to the baseline, while explanation quality improves by approximately 25%. Subsequent DPO alignment further enhances interpretability on the PreferenceSet. Cross-model validation on InternVL-3-8B and LLaVA-1.5-7B demonstrates the robustness and generalizability of the approach. The proposed framework improves detection of underrepresented mild damage cases, reduces high-risk misclassifications, and strengthens alignment between model reasoning and human judgment. Overall, it provides a reproducible pathway to develop reliable multimodal systems that deliver auditable, actionable disaster insights for emergency management.
Camila Blank, Zhuofan Ying, Christopher Potts +2cs.LG
Sycophantic agreement refers to a behavior in which language models excessively affirm the user, often at the cost of factual accuracy. Although sycophantic agreement is a well-known failure of model alignment, there is limited understanding of how it emerges from model training. In this work, we demonstrate that sycophantic agreement can emerge as an unintended consequence of widely used contrastive preference optimization objectives. Using the OLMo 3 post-training pipeline, we show that, for various pairs of teacher models across three families, there is a strong correlation between the log-ratio of the teacher model sycophantic agreement rates and the resulting student model sycophantic agreement rate. We further demonstrate that this unintended transfer is not limited to DPO but also occurs across 6 other preference optimization objectives. To understand whether this effect can be attributed to particular training examples, we analyze the preference data and find that the sycophancy signal is diffused across the entire dataset rather than concentrated in a sparse set of examples: each example appears neutral, i.e., there are no explicit instances of sycophantic agreement, and filtering based on probe-based data attribution or logit-linear selection fails to mitigate sycophancy without removing a large portion of the dataset. Overall, our findings suggest that the teacher models used to generate preference data can interact with alignment training objectives in unexpected ways, generalizing to undesirable and potentially harmful behaviors like sycophantic agreement.
Adapting large language models to user-specific preferences is often constrained by the cost of human annotation, making preference optimisation impractical in low-resource settings where preferences cannot be reliably labelled by LLMs themselves, e.g., due to cultural, subjective, or personalised contexts. In this paper, we investigate how language models encode preference information in their intermediate representations, finding that activations from chosen and rejected responses form distinct clusters across layers, even in pretrained models. Strikingly, this structure is strengthened by alignment on canonical datasets but erased when the target preferences differ from those the model was aligned on, suggesting aligned LLMs are poor judges for non-mainstream populations. Exploiting this structure, we propose training a lightweight linear probe on a few labelled preference pairs ($\leq$500) and using it to annotate large unlabelled datasets (50K+) for downstream preference optimisation. We systematically evaluate this approach across different datasets, preference optimisation methods and model scales and find that our method consistently outperforms direct training given the same annotation budget, and remains competitive against baselines trained on $50-100\times$ more labelled data in the majority of our settings. Code is available at https://github.com/alessioGalatolo/activ-pref-probe.
Productive dialogue alignment requires distinguishing \emph{surface coordination} (acknowledgments and smooth task progression) from \emph{epistemic alignment} (convergence of belief states); standard preference-based methods typically optimize response-level preferences without explicitly modeling the latter. We operationalize Theory-of-Mind (ToM) inference as a control signal within Frictive Policy Optimization by extracting, at each referring expression, a four-part belief structure: the speaker's intended referent, the addressee's interpretation, and each participant's model of the other's belief. This makes friction mechanically computable from epistemic-state comparisons, capturing \emph{silent divergence}, where both participants proceed confidently while grounding to different referents. We evaluate the signal at two levels. At the representation level, ablating the second-order channel reduces misunderstanding recall from $65\%$ to $26\%$. At the policy level, reward-shaping (FAR) and trust-region (FTR) variants improve intervention F1 and warranted-context calibration over DPO, with Brier scores independently supporting the calibration gains. Across three training runs, FAR and FTR remain substantially more stable, whereas DPO varies widely and can degrade intervention competence already present in the base policy. Thus, ToM-grounded friction provides a trainable signal for context-sensitive intervention under referential belief divergence.
Direct Preference Optimization (DPO) simplifies alignment through pairwise comparisons but assumes all observed preferences are reliable. Real data often violates this assumption, leading to reversed, weak, or ambiguous labels that cause harmful policy updates. To address this, we propose Posterior Label Correction DPO (PLC-DPO) to robustly optimize preferences by routing each pair's training signal as a clean, flip, or tie case. The key idea is to use the calibrated policy-reference margin as online evidence to take appropriate correction actions. This reframes noisy preference learning as actively correcting supervision direction and strength rather than merely filtering suspicious examples. Across 57 dataset-model-benchmark cells, PLC-DPO obtains the best mean win rate against DPO (60.5 vs. 55.5 for the next-best method). Injected-noise and tie stress tests, human disagreement analysis, and self-confirmation diagnostics further show that the routing remains stable and distinguishes flipped from weakly directional pairs.
Jinyoung Kim, Muhammad Khalifa, Lajanugen Logeswaran +4cs.CL
Natural-language critiques provide supervision beyond scalar rewards for non-verifiable generation, which lacks deterministic verifiers. In critique-guided refinement, a critic gives feedback on an initial response and an actor revises it. However, final revision quality does not reveal whether the critique was actually useful: a capable actor may improve without following the feedback, while valid feedback may fail if the actor cannot execute it. We frame critique as actor-conditioned revision guidance, where usefulness depends on whether the feedback helps the target actor address the intended weakness. We introduce TAIScore (Targeted Actionable Improvement Score), a reward that evaluates the instruction, initial response, critique, and revision together, assessing whether the critique targets a real weakness, whether the actor follows it, and whether the intended aspect improves. We use this reward to train an actor-tailored critic with GRPO, and use critique-guided refinements to construct DPO preference pairs for the actor, forming a co-evolving critic-actor loop where the critic adapts to the actor's changing capability. Experiments show that an 8B critic trained with TAIScore outperforms both a zero-shot 120B critic and critics trained with outcome-only or critique-only reward signals. Co-evolving the critic and actor further improves performance, suggesting that effective critique supervision should adapt as the actor changes.
Preference optimization is widely used to align large language models with human preferences, but preference-data composition may also influence privacy-relevant memorization. We examine whether adding synthetic privacy-preference pairs to Direct Preference Optimization (DPO) is associated with lower canary-based memorization signals without modifying the objective or introducing a formal privacy mechanism. We propose Privacy-Pressure Preference Mixing (P3M), a data-composition protocol that varies the amount of privacy-preference data while keeping helpfulness and harmlessness preference data fixed. We evaluate a non-privacy Baseline and privacy-mixing ratios of 0.5, 1.0, and 2.0 using Gemma 3 270M-IT across five random seeds and validate the same four conditions using 4-bit-quantized Gemma 2 2B-IT across three seeds. Overall, under the tested conditions, privacy-preference mixing is associated with lower mean canary suffix log-likelihood proxy values across both model settings and lower aggregate membership-inference attack performance relative to the Baseline in the mixed-source 2B evaluation. Specifically, across the privacy-aware 2B configurations, the mean area under the receiver operating characteristic curve (AUROC) ranges from 0.596 to 0.629, and the mean area under the precision-recall curve (AUPRC) ranges from 0.541 to 0.575, compared with 0.804 and 0.790, respectively, for the Baseline. However, the reduction in membership distinguishability does not hold uniformly across data sources. Moreover, the relationship between the privacy ratio and harmlessness preference accuracy varies by model setting, whereas helpfulness preference accuracy remains broadly stable. These findings suggest that P3M should be viewed as a lightweight empirical protocol for examining privacy-utility-safety trade-offs rather than as a formal privacy guarantee or a defense against extraction attacks.
Interactive dialogue games test a capability that static benchmarks largely leave implicit: a model must carry state across turns, interpret feedback, and choose valid actions under changing constraints. We study this setting in the LM Playschool Challenge with a 2B open-weight model, and find that many failures are not only broad knowledge failures but also local decision failures: repeated guesses, malformed actions, and violations of feedback that the model has just seen. These diagnostics motivate a training recipe organized around three steps: acquire broad game participation through supervised fine-tuning, repair mechanically verifiable failures within one targeted dialogue-game family using turn-local preference pairs, and preserve general capabilities beyond these dialogue games. In the official final evaluation, our submission improves public clemscore from 10.67 to 38.92 and closed in-domain score from 13.41 to 41.17, while approximately preserving aggregate static performance (44.14 vs. 44.24 for the baseline). Out-of-domain clemscore remains low at 7.88, with the largest gains concentrated in unseen variants of the targeted family. Our results suggest that broad SFT brings most of the model's capability improvement; turn-local supervision can be effective when failure detection is precise, with observed transfer concentrated primarily within-family.
Direct Preference Optimization (DPO) is a widely used objective for aligning language models from preference data, with the coefficient $β$ commonly interpreted as controlling the KL constraint to a reference policy. We show that $β$ entangles two distinct roles: it governs the effective inverse preference-noise scale and simultaneously rescales the optimization dynamics, coupling this scale with the effective step size. As a consequence, at a fixed learning rate the achieved policy deviation is non-monotone in $β$: it vanishes in a dead zone at small $β$, reaches a peak at an intermediate value, and decreases again for larger $β$. Moreover, standard DPO loss values are not comparable across $β$: runs with nearly identical loss curves can differ several-fold in KL divergence from the reference model. This entanglement obscures the role of $β$, increases sensitivity to hyperparameter choices, and complicates learning-rate scheduling. We propose a centered-softplus reformulation that is argmin-equivalent to DPO for $β>0$, while making the inverse preference-noise-scale and learning-rate effects explicit and independently tunable. The normalized centered-softplus objective also admits a continuous $β\to0$ endpoint that reduces to a linear preference-margin objective.
Mathematical reasoning has seen rapid progress in large language models (LLMs), yet existing methods optimize predominantly for final-answer correctness, raising the question whether models truly internalize mathematical concepts or merely memorize solution patterns. In human mathematics education, example-based reasoning such as constructing counterexamples to test theorem boundaries reflects deep conceptual understanding, but remains underdeveloped in current LLMs. Enhancing this capability through preference optimization presents two key challenges: (1) the model's limited example-based reasoning ability makes constructing effective preference pairs inherently difficult; and (2) capability acquisition is progressive, as the model must first learn to adopt this strategy before learning to apply it correctly. Therefore we propose INSPIRE, an Internalize-Then-Improve approach combining Reference-Guided Student Internalization (RGSI), which produces high-quality preference candidates under the policy model's own distribution, with a stage-wise rubric preference training strategy that decomposes learning into method-oriented and correctness-oriented stages. Experiments across multiple model scales and families demonstrate consistent improvements, even surpassing larger open-source models, while evaluations on out-of-distribution benchmarks confirm no degradation in general mathematical reasoning ability.
Dongyue Li, Ziniu Zhang, Lu Wang +1cs.LG cs.AI cs.CL
We study learning a mixture of $k$ Plackett-Luce models from multi-way ranking responses from annotators that may represent heterogeneous underlying preferences. This problem has many applications in AI alignment and preference optimization. Prior work has studied mixtures of Bradley-Terry models from pairwise comparisons. However, estimating a mixture of multi-way ranking models can become theoretically unidentifiable when $k$ exceeds $m/2$, where $m$ is the ranking length. We design an efficient algorithm to address this issue by first augmenting the rankings to a larger size (e.g., generating comparisons from a base model), followed by a gradient-based estimation to reduce inference cost (in the input embedding space). With this procedure in mind, we then fit a mixture of Plackett-Luce (PL) models via an expectation-maximization-style iteration, or MoPLEx in short. We conduct extensive experiments to verify this algorithm. First, we find that the gradient-based approximation estimates true probabilities with less than 5% error on models with up to 34 billion parameters. Second, MoPLEx improves clustering and ranking accuracy by an average of 43.7% and 15.2% over baselines using a single PL model or a mixture of Bradley-Terry models, on UltraFeedback and PERSONA datasets. These results demonstrate the effectiveness of MoPLEx for tackling multi-way rankings following heterogeneous preferences through measuring alignment via gradients.
Arthur Corrêa, Paulo Nascimento, Samuel Monizcs.LG
Multi-task vehicle routing problem (VRP) solvers seek to handle multiple VRP variants within a single unified model, avoiding the need to train a separate model for every variant. In spite of recent progress, current approaches remain limited on two fronts. On the training side, reinforcement learning suffers from reward-scale disparities and shrinking advantage signals as policies improve, whereas preference optimization stagnates once sampled tours become near-identical and thus fundamentally limited by the quality of the policy's own generated solutions, leaving both paradigms with weak supervision as training progresses. On the architecture side, existing fully shared encoders entangle constraint-dependent representations across heterogeneous variants, which limits generalization. We address these gaps with two model-agnostic contributions. First, we propose Preference Optimization with Locally Augmented Refinement (POLAR), a novel training algorithm that applies a local search refinement pass to the best decoded tour before forming preference pairs, yielding much more informative pairwise margins. Second, a Progressive Layered Extraction (PLE) encoder routes each encoder layer through one shared expert and a set of task-specific experts via a gating mechanism, progressively separating common routing structure from constraint-specific encodings. Through extensive experiments on various VRP variants, we show that POLAR and PLE together elevate the current state-of-the-art among neural multi-task solvers. We reduce the average gap to reference solutions by 21.3% relative to the strongest published baseline on 16 in-distribution variants, and outperform prior neural methods on 27 out of 32 unseen variants. Ablation studies confirm the efficacy of each contribution, showing that both improve cross-problem generalization across multiple backbone model architectures.
Aligning large language models to human preferences is crucial for real-world deployment but frequently incurs an alignment tax, leading to the catastrophic forgetting of pre-trained general capabilities. While previous works primarily frame this problem as an optimization or architectural challenge, the inherent characteristics of preference data that drive this degradation remain largely underexplored. In this paper, we propose BALIGN, a balanced data selection strategy that explicitly mitigates catastrophic forgetting while optimizing alignment efficacy. Through theoretical and empirical analyses of the preference optimization gradient, we identify three key data-centric features that dictate parameter drift: the reference model's log-probability margin, the token length difference between chosen and rejected responses, and the TF-IDF similarity to general capability corpora. By aggregating these orthogonal features into a unified composite risk score, BALIGN systematically filters out high-risk preference samples that disrupt intrinsic model parameters or provide minimal alignment utility. Extensive experiments on standard human preference datasets demonstrate that BALIGN strongly preserves foundational capabilities without compromising alignment gains, consistently achieving the optimal Pareto frontier with minimal computational overhead.
Non-verbal vocalizations (NVs), such as laughter, coughs, and sighs, are essential for expressive TTS, but the effectiveness of preference optimization for NV generation remains poorly understood. We systematically study preference optimization for NV-capable TTS, focusing on preference signals, preference-pair construction, and DPO-based optimization objectives. We formulate an NV-aware character error rate (NV-CER) by treating NV tags as distinct output symbols and computing a weighted pinyin-based CER over both verbal and non-verbal content, enabling controllable optimization of NV realization without modifying the underlying optimization algorithm. Experiments on Emilia-NV and the augmented NV-Bench covering 18 NV types reveal how different design choices affect NV realization and lexical fidelity, and establish an effective setup using standard DPO. Objective, LLM-based, and human evaluations provide converging evidence for our findings, offering practical insights into NV-aware post-training for expressive TTS.
Seungyoon Lee, Minhyuk Kim, Jungseob Lee +1cs.CL cs.AI
The alignment of Large Language Models heavily relies on English-centric high-quality preference data, which often leads to suboptimal performance in other languages. In this paper, we propose Cross-Lingual Ranking Preference Optimization (CRPO), a novel framework that leverages robust preference knowledge from English to facilitate preference alignment in the target language. We design a hierarchical structure within parallel preference pairs across the target language and English to jointly optimize intra- and inter-lingual preferences, thereby enhancing language adaptation and output quality. Building on the LambdaLoss framework, CRPO goes beyond the binary comparison based optimization by providing a relative ranking signal across multiple candidate responses. Our experiments across five languages with varying resource scales demonstrate that CRPO consistently outperforms standard approaches in both instruction-following and knowledge utilization capability. Notably, the robust performance gains observed across various weighting schemes further validate the empirical effectiveness of our hierarchical design in a multilingual setup. Furthermore, our findings highlight that CRPO significantly improves both reward margins and the log-probability of desirable responses, contributing to a more stable preference manifold for cross-lingual alignment.
Iterative preference optimization is essential for aligning Large Language Models on mathematical reasoning tasks, yet its efficiency is often throttled by signal scarcity: as the model improves, static problem sets become increasingly mismatched to the model's evolving competence, producing rollouts that are either too easy or too hard and therefore non-informative, which leads to a scarcity of valid preference pairs. We propose DIAG, a Diagnostic Iterative Alignment and Generation framework that adaptively reshapes the practice distribution to increase informative supervision and focus training near the student's current competence boundary. DIAG consists of two phases: (1) diagnosing valid preference-pair yield to calibrate the exploration-exploitation trade-off and allocate topic quotas via an Empirical Bayes shrinkage estimator, thereby prioritizing high-yield concepts; and (2) generating targeted practice, where a teacher synthesizes variants from the student's failure traces. We further provide a theoretical view interpreting DIAG as a teacher-mediated approximation to KL-regularized reweighting of the practice distribution toward the student's competence boundary, where valid preference-pair yield is maximized. Experiments show that DIAG boosts yield across iterations and delivers stronger reasoning performance under an iso-effective training budget, demonstrating that it can distill more informative preference supervision for mathematical reasoning.
Text-to-SQL aims to translate natural language questions into executable SQL queries over relational databases, requiring multi-stage structured reasoning over database schemas and query constraints. However, existing methods treat this task as single-step generation, where models optimize entire SQL sequences without targeted feedback at key decision points and lack support for interacting with and controlling the intermediate generation process. To address this issue, we propose SPOC-SQL, which decomposes Text-to-SQL into four sequential subtasks following standard SQL execution logic and designs stage-specific optimization strategies for the model to learn key decisions. Specifically, we propose the implementation of fine-grained preference optimisation at key decision points across SQL stages, with the objective of enhancing structured decision-making during query construction. Furthermore, a structured decomposition strategy is designed, facilitating stage-wise intervention and correction through explicit intermediate representations. This results in more controllable and reliable SQL generation. Experiments demonstrate that incorporating stage-wise human knowledge consistently improves performance, validating the effectiveness of stage perception controllable generation.
Idris Nechnech, Sehwan Kim, Jimin Seo +4cs.AI cs.SE
Process supervision has improved mathematical reasoning, where intermediate steps are naturally expressed as chains of thought. In code generation, however, process supervision remains underexplored because there is no standard notion of a step. Supervision can target lines, reasoning traces, or program states, making it unclear what to label and optimize. We propose STEP-KTODER, a framework for code preference optimization that defines steps as module-level functions in decomposed multi-function programs and assigns binary correctness labels via automatically generated unit tests. Our method provides a code-specific instantiation of stepwise KTO, combining function-level process supervision with outcome-level feedback on the full program. We evaluate on HumanEval(+), MBPP(+), BigCodeBench, and LiveCodeBench, showing that STEP-KTODER improves over outcome-only KTO and DPO. Further analysis shows that execution-based labels are essential: LLM-as-a-judge annotations systematically over-predict function failures, corrupt positive step labels, and degrade downstream preference optimization. Code is available at: https://github.com/inechnech/STEP-KTODER.
Preference optimization is a standard alignment method for generative models, yet extending it to continuous-time dynamics remains non-trivial. In flow matching, reward-driven updates modify transport trajectories without an inherent constraint to the pretrained data manifold and can move terminal samples off the pretrained support. We formalize this failure mode as manifold drift. Theoretically, we show that optimal flow matching recovers the terminal data distribution, whereas a preference update leaves the pretrained manifold whenever its induced terminal displacement has a nonzero normal component. As a remedy, we propose ThermoDPO, a temperature-controlled objective that anchors pairwise preference optimization on preferred samples. Across temperature regimes, this objective connects rejection sampling fine-tuning and FlowDPO and controls a pointwise reconstruction-based surrogate for manifold distance. To counteract diminished signals at low temperatures, we further introduce a weighted variant, ThermoDPO-weighted. On the main toy benchmark, ThermoDPO-weighted attains a StrictScore of 0.899, compared with 0.629 for FlowDPO and 0.857 for FlowDPO+RFT. On SD3.5-M at CFG = 4.5, it improves OCR by 47.5% and the average of four metrics by 16.0%.
Cyclic peptides are emerging as promising molecular scaffolds in drug discovery due to their high binding affinity and structural stability. However, extending generative models from linear to cyclic peptide design remains challenging, as cyclization sharply restricts the feasible design space through coupled geometric and biophysical constraints. Moreover, limited training data has led existing approaches to rely largely on zero-shot generation or post hoc filtering, resulting in low yields of feasible designs and limited control over multi-objective trade-offs. To address these limitations, we propose FAR-DPO (Feasibility-Aware and Robust Direct Preference Optimization), an architecture-agnostic framework that steers generative models toward structurally and biophysically feasible cyclic peptide designs, particularly for challenging targets. FAR-DPO integrates feasibility-aware preference construction with difficulty-aware group-robust optimization. Specifically, it constructs within-target preference pairs through feasibility-gated multi-objective dominance and adaptively reweights predefined difficulty groups according to their current preference losses. On the CPSea LNR benchmark, under a fixed generation budget, FAR-DPO increases overall success rate from 46.89% to 57.79% on PepGLAD and from 47.96% to 49.57% on PepFlow. These gains also extend to the hardest target quartile and are accompanied by more favorable best-per-target binding scores. Together, these results demonstrate FAR-DPO's effectiveness in improving feasibility and target-wise robustness.
Direct Preference Optimization (DPO) has emerged as an effective approach for aligning large language models (LLMs) with human preferences. However, its adaptation to multimodal settings remains unexplored. Through representational analysis, we identify a key limitation in multimodal preference optimization, which we term visual insensitivity: models often fail to distinguish between images and those with critical visual context removed. Our theoretical analysis further uncovers two manifestations of this problem, namely Across-Image Insensitivity and Within-Image Insensitivity. To address these challenges, we propose Perception-Enhanced Alignment DPO (PEA-DPO), a framework for multimodal LLMs alignment, which explicitly leverages visual preference signals to overcome visual insensitivity. We further provide a theoretical analysis demonstrating that PEA-DPO provably mitigates both failure modes. Empirical results demonstrate that PEA-DPO enhances sensitivity to visual context while preserving the language modeling capacity of the base model. Evaluations across three hallucination benchmarks using MLLMs of varying scales show that PEA-DPO effectively mitigates visual insensitivity, achieves stronger multimodal alignment, and substantially reduces hallucinations.
Infrared (IR) spectra provide characteristic signals of molecular structure, which are often interpreted by experts via functional-group identification or library matching, making the process time-consuming and ambiguous. Recent machine learning methods have made progress in molecular structure elucidation using molecular formulas and IR spectra. However, these models often infer unreasonable candidate molecular structures, including top-ranked predictions. More specifically, the molecular formula implied by a candidate structure often fails to match the input molecular formula, and the candidate's theoretical IR spectrum is often inconsistent with the observed IR spectrum. To address these issues, we propose Formula- and IR-Matched Preference Optimization (FIRMPO), a general and plug-and-play chemical feedback-driven preference optimization framework for molecular structure elucidation. FIRMPO incorporates chemical feedback as preference signals based on exact molecular formula matching and IR spectral consistency to guide reasonable structure predictions. Unlike generic preference optimization methods, FIRMPO is tailored to molecular structure elucidation while remaining model-agnostic, enabling it to be readily integrated with different structure prediction models in this class. This encourages models to prioritize structures that satisfy the chemical feedback, leading to a substantial improvement in the accuracy of top-ranked predictions. Extensive experiments on three widely used IR datasets show that FIRMPO significantly improves molecular structure elucidation accuracy over existing baselines.
Tony Tu, Sayan Chakraborty, Ruomeng Xu +2cs.AI cs.CL cs.LG
Aligning a language agent to several objectives at once is a persistent failure mode of preference-based training: when objectives are combined additively, optimization collapses onto whichever is cheapest to improve and sacrifices the rest, so a support agent learns to sound warm while giving no real help. The root issue is that an additive reward has no notion of balance. We introduce Mint (MIN-selection preference disTillation), a one-line change to preference distillation: rather than ranking sampled candidates by a weighted sum of rewards, we rank them by their weakest objective, distilling the best-balanced candidate over the most lopsided one with an unchanged DPO objective. This is the p -> negative infinity limit of a generalized-mean family spanning additive to worst-case selection. Across cooperative emotional support and adversarial negotiation, min-selection lifts both objectives while sharply cutting their imbalance; on emotional support it raises the weaker axis from 0.37 to 0.64 (p < 10^-40), surpassing human experts and persisting across full multi-turn rollouts. A turn-by-turn analysis yields our central finding: min-selection corrects imbalance in proportion to how imbalanced the reference policy is, and its benefit endures over an interaction precisely as long as that imbalance does.
Large language models (LLMs) remain vulnerable to harmful requests and jailbreak attacks. Parameter-efficient safety alignment methods based on prompt tuning typically rely on a single global prompt or externally selected prompt modules. Such static designs struggle to maintain a cross-category safety boundary while generating constructive responses tailored to specific risks and avoiding over-refusal of benign inputs. To address these limitations, we propose HiRoute, an input-adaptive hierarchical prompt-tuning framework that separates category-agnostic safety control from category-specific response guidance. HiRoute first trains a lightweight hierarchical router on representations extracted from a frozen LLM to jointly detect harmful intent and predict multi-label risk scores. It then freezes both the backbone model and the router and uses preference optimization with alternating gradient updates to learn a shared coarse-grained prompt and a set of fine-grained prompt experts as continuous embeddings. At inference time, benign inputs bypass the safety branch, whereas risky inputs are processed using the shared prompt together with a router-weighted mixture of risk-specific prompt experts. Experiments across three instruction-tuned models show that HiRoute achieves high safety rates across multiple safety benchmarks while preserving safe-response helpfulness, reducing over-refusal, and maintaining competitive performance on general-purpose tasks.
Yuanyu Li, Jintao Xu, Zijiang Liu +6stat.ML cs.AI cs.LG math.OC
Neural combinatorial optimization (NCO) relies on parallel solution sampling for training, yet existing methods fail to fully exploit the rich information latent in a co-sampled solution group. Preference-optimization methods anchor on the single best solution and discard fine-grained quality and structural signal from all other peers-a failure we term gradient signal polarization. Mean-based baselines instead weight peers uniformly, so structurally near-identical peers flood the baseline with redundant information and keep gradient variance high-a failure we term baseline redundancy. We propose SSPO (Structure-Aware Similarity-Weighted Preference Optimization), which scores all $B$ sampled solutions jointly through a dissimilarity-weighted leave-one-out baseline: structurally distinct peers receive higher weight, resolving both failures in a single mechanism. The baseline uses zero-parameter, problem-adaptive solution embeddings built from the encoder's existing node representations. Experiments on TSP, EFL, and JSP benchmarks show consistent gains over prior best-anchor and uniform-weight baselines. A direct comparison against uniform RLOO on TSP and EFL confirms that structure-aware weighting is the primary driver of improvement. The SSPO-trained EFL policy has been deployed in a production facility-location system at JD$\mathord{.}$com, confirming practical viability at scale.
Multimodal large language models (MLLMs) have made rapid progress, yet they still exhibit object hallucination, generating plausible but incorrect descriptions that are inconsistent with the visual input. Direct Preference Optimization (DPO) mitigates this by training models to prefer non-hallucinated responses over hallucinated ones, and recent efforts further enrich the preference data with relevant context. However, it remains unclear whether DPO actually leverages such context. To investigate this, we propose Contextual Preference Gain (CPG), a simple metric that measures how much a model's preference strengthens when relevant context is provided. We find that higher CPG consistently corresponds to lower hallucination, yet standard DPO and its variants exhibit only limited CPG, indicating that they underutilize contextual information and thus remain prone to hallucination. To address this, we propose Context-Calibrated DPO (C$^2$-DPO), which directly maximizes CPG while preserving the original preference ordering. Across multiple benchmarks, C$^2$-DPO substantially reduces hallucination without compromising general reasoning, relatively reducing the Object HalBench hallucination rate of Qwen2-VL-Instruct-2B by 36%. Code is available at https://github.com/mlvlab/C2-DPO