Historical people may appear under different languages, scripts, and transcription traditions, while distinct individuals may share highly similar or even identical names. This makes historical identity reconciliation more than a problem of string matching or transliteration. We introduce MHER, a provenance-controlled benchmark for pairwise reconciliation of person-name attestations from the Mongol world. MHER contains a balanced 396-pair Name-only core over 84 primary historical persons and a stricter 160-pair Source-grounded subset constructed from mention-by-source evidence, with entity-disjoint development and test splits. Across five generative systems, correctly Source-grounded evidence improves paired TEST accuracy by 12.96 to 94.44 percentage points relative to Name-only input. On five identical-surface different-person cases, all models fail under names alone (0/25 model-item decisions), whereas Source-grounded evidence yields 24/25 correct resolutions, with the remaining output an abstention. Context-only ablations show that historical descriptions often carry substantial identity information, while explicitly signaled misgrounding controls produce substantially lower performance. We also find that names are not uniformly beneficial: for Qwen3-8B, restoring surface forms converts ten otherwise correct Context-only distinctions into false identity merges. These results show that historical entity reconciliation depends not only on surface correspondence, but on whether identity judgments respond appropriately to provenance-controlled historical evidence. MHER therefore provides a controlled framework for studying evidence use, abstention, and failure modes in historical NLP.
Assine o Pro, sem anúncios
Faça upgrade para uma leitura sem interrupções e acesso prioritário a novas fontes.
Reasoning-Induced Misalignment, where fine-tuning on reasoning data containing no harmful content, including mathematics, code, and problem-solving with chain-of-thought traces can induce harmful behaviors of LLM, posing a serious challenge to the safety of LLM reasoning. Cross-architecture, cross-scale, and cross-dataset checks show that RIM does not always emerge. Previous work attributed RIM to neuron-level entanglement, but did not identify the geometry of the representation space underlying this entanglement or propose a training-time fix. We provide both: a representation-space analysis of RIM and the Safety-Direction Penalty (SDP), which penalizes movement along a learned safety direction during reasoning fine-tuning. The analysis extracts two activation-space directions, one encoding reasoning ability and the other safety behavior. These directions are coupled: fine-tuning that improves reasoning shifts safety representations, and prompts with larger shifts show larger safety degradation. CKA distance ratios and probes locate the safety-decision layers where this shift is most relevant. These findings guide the design of SDP: the coupling motivates penalizing displacement along the safety direction, and the layer localization sets the initial scope. When the initial scope leaves compensatory shifts beyond the penalized layers, the same diagnostics guide iterative expansion. On Qwen2.5-3B and 7B, SDP restores safety while preserving benchmark reasoning performance.
Self-reflection is a powerful mechanism for credit assignment in human learning, converting sparse outcome feedback into actionable guidance. However, its potential for post-training Large Language Models (LLMs) remains underexplored. We propose Self-Reflective Policy Optimization (SRPO), a framework that internalizes this capability. SRPO enables LLMs to analyze their own completed trajectories, synthesize errors into concise "reflection patches," and use reflection-conditioned teacher scores on student on-policy rollouts as dense token-level training signals. This process effectively transforms sparse terminal supervision into dense, token-level learning signals without requiring external critics, separate reward models, or larger teacher models. We demonstrate that SRPO achieves state-of-the-art performance across mathematical reasoning and long-horizon agentic benchmarks with exceptional data efficiency. Using a Qwen3-8B base model, SRPO attains 73.3% on AIME'24 using only 8% (0.08x) of the training FLOPs required by scaled supervised fine-tuning, while significantly improving success rates on WebShop (64.7%), ALFWorld (76.8%), and SWE-Bench-Lite (31.2%). Code is available at https://github.com/Galleons2029/SRPO
We present Team Semiintelligencn's solution for the ACM RecSys 2026 TalkPlayData Challenge, addressing conversational music recommendation through a multi-modal and personalized conversational recommender system. Our submitted system employs a three-stage pipeline: (1) multi-modal retrieval constructing decay-weighted centroids across seven dense embedding spaces - track- and user-level CF-BPR, Qwen3 (metadata, lyrics, attributes), CLAP audio, and SigLIP visual - supplemented by BM25 lexical retrieval and an artist substring-match signal, all fused via weighted Reciprocal Rank Fusion (RRF) with optimized signal weights; (2) lightweight reranking (history filtering, popularity smoothing, and catalog diversity penalization); and (3) persona-diversified response generation using GPT-4o-mini. Beyond this submitted configuration, we report development-time experiments with additional components - constrained LLM-guided artist injection, album continuation signals, XGBoost LambdaMART, and a superior GPT-4.1 response prompt - that were not deployed to Blind B due to cost and complexity constraints. We optimize RRF weights on a 500-session development split via differential evolution, improving MRR by +19.5%. On Blind A, we observe that unconstrained LLM-guided injection across 54 sessions causes catastrophic nDCG regression (-18.9%), while conservative injection on only 9 sessions yields the best observed Blind A nDCG - a finding we present as a Blind A observation warranting further validation. The submitted system achieves a Blind B composite score of 0.3213.
Timely risk classification is essential in many clinical monitoring settings, where decisions must balance the benefit of classifying patients early for subsequent intervention against the value of observing additional data. Yet most existing statistical and machine-learning methods are designed for fully observed trajectories and offer limited control over key operating characteristics such as sensitivity, specificity, and monitoring cost. We cast the sequential classification problem within a multi-objective optimization framework targeting these three criteria. We characterize the optimal decision rule through a value recursion that quantifies, at each time point, the trade-off between immediate classification and continued monitoring. To estimate the rule from data, we formulate a constrained optimization problem that maximizes specificity while enforcing prespecified sensitivity and monitoring-cost constraints. We then develop an estimation procedure that employs a recurrent neural network to approximate the evolving value processes and a primal--dual updating scheme to satisfy the performance constraints. Through simulation studies and an application to continuous glucose monitoring for hypoglycemia risk prediction, we demonstrate that the proposed method yields accurate and timely sequential decision rules that adhere to the desired operating characteristics.
Vision-Language-Action (VLA) models can turn multimodal context into robot actions, but their action decoders are still trained largely by behavior cloning. This supervises which motor command was demonstrated while leaving implicit the local objective served by the behavior under the instruction. Future-based supervision enriches action learning with frames, latent observations, trajectories, or motion representations, but these signals capture particular realizations of what may happen rather than the shared semantic objective of the forthcoming behavior. We propose Intention Distillation (INDI), which distills behavior-level intent into the action decoder. During training, a frozen teacher VLM interprets a demonstrated segment from the current observation, instruction, coarse action summary, and corresponding execution video. From its standard inputs, the deployed VLA recovers the resulting multimodal intent representation at an intermediate decoder layer and uses it to organize action prediction together with representations of how the behavior unfolds and what it achieves. On SimplerEnv-Bridge, INDI improves GR00T-N1.7 from 64.3% to 84.7%, and on RoboCasa Kitchen it improves the controlled GR00T-N1.7 baseline from 64.1% to 70.3%, with consistent gains on π_{0.5} across both benchmarks. In real-world tasks, INDI improves average success from 62.0% to 68.7%, with gains of up to 12.0 pp on longer-horizon tasks. Further analyses show that the recovered latent is used by the decoder, captures behavior objective and execution progress, and organizes downstream predictions in an objective-dependent manner. These results show that action decoders benefit from explicitly modeling the semantic objective of the behavior they generate.
Narrow fine-tuning on small, domain-specific datasets can produce broad and surprising changes in model behavior-a phenomenon called weird generalization (WG). Yet, it remains unclear what features of the fine-tuning data are necessary for WG to arise. Here, we address this question by investigating a range of plausibly relevant features, including dataset size, composition, language, presentation style, and novelty relative to a model's parametric knowledge. Further, since WG evaluations rely on small question sets that assess the extent of the generalization, we also analyze how sensitive this measurement is to the set of questions used. Experiments with three open-weight models on four datasets show that the degree of WG (1) depends heavily on dataset composition and language (more than on size); (2) is greater for data familiar from pretraining than for novel data; and (3) is sensitive to the set of evaluation questions used. Collectively, these results indicate that WG is a product of quite fragile properties of both training and evaluation data. As such, we argue that WG is more plausible as an adversarial threat-requiring careful data engineering-rather than as a significant hazard inherent to routine fine-tuning.
As large language models are increasingly used in data-scarce and evolving task scenarios, few-shot in-context learning (ICL) has become a key paradigm for task adaptation. However, direct ICL often uses a small set of examples without explicitly abstracting task rules, making it sensitive to example construction. In contrast, human learners often reduce such sensitivity by first summarizing task rules from examples and then applying them to new instances. To evaluate this ability, we propose StrategyBench, which selects strategy-inducible tasks from BIG-Bench, constructs reference strategies, and defines evaluation metrics along two dimensions: strategy quality and downstream utility. We further analyze strategy induction from three perspectives: task variation, model configuration, and adaptation setting, covering category-wise differences, generator-executor choices, demonstration design, and SFT-based adaptation. Experiments show that explicit strategy utility differs substantially across task categories and depends on both strategy generation and execution conditions. The benchmark is released at: https://anonymous.4open.science/r/StrategyBench-D53C.
Vision-language models (VLMs) achieve strong performance on video and image-sequence benchmarks, yet it remains unclear whether they capture temporal structure. To study this question, we formulate temporal grounding as an anomaly detection problem, providing a simple and controlled evaluation that directly tests sensitivity to temporal consistency. We introduce TimeCatch, where temporal anomalies are created by swapping consecutive frames and frame-level anomalies by replacing a frame with Gaussian noise. Models are evaluated on anomaly detection and localization tasks across four synthetic and real-world datasets, alongside a human study. Our evaluation reveals a substantial gap between frame-level and temporal anomaly detection. While VLMs consistently detect frame-level anomalies and often localize them accurately, they perform near chance on temporal anomaly detection and only modestly above chance on localization. Humans, in contrast, achieve near-ceiling performance on both tasks. Additional analyses across model scales, prompting strategies, sequence lengths, and visual similarity suggest that these failures cannot be explained solely by limitations in perception or model capacity. Together, these findings indicate that current VLMs can identify anomalies within individual frames but struggle to integrate information across frames to reason about temporal consistency. TimeCatch provides a controlled benchmark for evaluating temporal grounding in vision-language models.
Time series forecasting (TSF) is evolving toward multimodal and agentic settings, yet using foundation models remains uneconomical in resource-constrained scenarios, where compact, specialized forecasters are more desirable. However, lightweight forecasters typically require substantial training data, limiting their use in domains with scarce, slowly accumulated, or privacy-sensitive time series. To address this dilemma, we investigate the challenging problem of few-shot learning for lightweight forecasters. We propose MetaCaster, a meta-harness-optimized multi-agent framework that uses agentic data generation to automatically train specialized lightweight forecasters from only a few examples and textual contexts. Our work highlights a new TSF paradigm in which agents act not as forecasters but as intermediary engineers that prepare efficient, task-specific forecasters for deployment. Experiments on 18 datasets, 23 state-of-the-art lightweight forecasters, and 14 baselines demonstrate that MetaCaster achieves both data efficiency and computational efficiency while maintaining high-quality TSF performance.
Memory is becoming a default subsystem in deployed LLM agents to provide persistent personalization and continuity. This naturally prompts a question: will memory system introduce new vulnerabilities into agents? Thus we propose InjecMEM, a novel memory injection attack paradigm that requires only a single interaction (no read/edit access to memory store) to steer later responses of related queries toward a pre-specified output. Guided by the retrieval-then-generate mechanism of memory systems, we craft the injection with a retriever-agnostic anchor and an adversarial command. The anchor contains high-recall topical cues so that downstream retrieval consistently associates the record with the target topic. The command is a short sequence optimized to remain effective under uncertain fused contexts, variable placements, and long prompts so that it reliably steers outputs once retrieved. We learn the command via gradient-based coordinate search, averaging over synthetic prompt templates and insertion positions, and extend it to joint optimization across backbones to study transfer. Evaluated across multiple memory systems and backbone models, InjecMEM achieves reliable topic-conditioned retrieval and targeted generation, remains effective under memory drift, and leaves non-target queries unaffected. Our results underscore the need to harden memory systems and provide a reproducible framework for studying agent memory.
A machine learning-assisted framework for the inverse design of pixelated millimetre-wave patch antennas targeting the 22--30 GHz band is presented. The antenna surface is represented as a 19x23 binary pixel grid on a Rogers RT/duroid 5880 substrate, where each pixel is either metal or empty, with a continuous electrical path from the feed enforced by design. An initial dataset of approximately 6,000 full-wave CST simulations was collected from structured random pixel patterns, of which only around 40% achieved a resonance with |S11| <= -10 dB anywhere in the band, resulting in an imbalanced dataset. To improve simulation efficiency, an XGBoost binary classifier was trained on this data to distinguish resonant from non-resonant patterns before simulation. Using the classifier as a pre-simulation filter, an additional 4,000 patterns were selected and simulated, raising the overall proportion of resonant designs in the combined 10,000-sample dataset from approximately 40% to 52%. A hybrid CNN-BiLSTM forward surrogate was then trained on this augmented dataset to predict the full complex S11 response across 801 frequency points, using a physics-guided composite loss that explicitly emphasises resonance dip accuracy. Finally, an inverse design model was developed that optimises in a compact 64-dimensional latent space using gradient descent to generate pixel patterns matching a desired S11 specification. The results show good agreement between the surrogate-predicted and CST-simulated |S11| responses for the generated designs and demonstrate the feasibility of automatically designing and reconfiguring antenna structures.
Anomaly detection is often applied to data stored in relational databases, yet most existing methods require flattening multiple tables into a single feature matrix. This flattening can obscure entity identity, schema structure, and multi-hop dependencies, limiting the detection of anomalies that depend on relational context rather than isolated feature values. Beyond preserving relational structure, relational anomaly detection raises an additional challenge: how to incorporate symbolic behavioral evidence into learned relational representations. To address these challenges, we study relational anomaly detection, where the goal is to identify anomalous entities or events in a multi-table database. We propose RAD, a rule-augmented relational anomaly detector that combines heterogeneous graph representation learning with refined symbolic rule signals. RAD derives candidate rules from random-forest paths over flattened summaries of the entities or events being scored, refines them into compact interpretable predicates, injects the resulting rule features into the graph model, and learns anomaly scores using reconstruction-based and pairwise-ranking supervision. To evaluate this setting, we introduce a relational anomaly detection benchmark spanning three settings: LANL cybersecurity event detection and two unexpected user-churn anomaly tasks derived from Amazon and H&M relational databases. Experiments show that RAD improves anomaly ranking over flattened tabular detectors and relational baselines under natural class imbalance, achieving the best average rank on AUROC and AUPRC across the benchmark. Ablations show that direct rule injection and ranking-based supervision are key contributors to performance, while edge reconstruction is not uniformly beneficial. Our code and data are available at: https://github.com/noahd15/RAD_RelationalAnomalyDetection.
The quadratic growth of attention computation and key-value (KV) cache with respect to sequence length is a central bottleneck for ultra-long-context language models and high-resolution generative models. We propose ProxyFormer, a general dual-stream architecture built upon proxy tokens. In each layer, fine-grained local features are compressed bottom-up into a small set of proxy states; expensive global interactions are performed only in the compressed proxy space; the globally contextualized proxies are then decompressed and injected top-down back into the local stream. Because the local stream persists across layers, fine-grained information that is not captured by one compression step remains accessible for later refinement, alleviating the irreversible information loss of conventional one-shot compression. We further introduce factorized multi-level compression/decompression, layer-wise dynamic compression ratios, asymmetric dual embeddings, and a proxy-only KV-cache inference scheme. On a 16GB GPU with batch size 1, a standard decoder-only model can train sequences of only about 20K tokens, whereas ProxyFormer with a compression ratio of 64 extends the trainable sequence length to about 0.7M. A model trained with a 64K window retains 92%-95% retrieval accuracy on a multi-needle retrieval task with 1,048,576 tokens, and a model trained with an 8K window exceeds 94% accuracy when extrapolated to 256K tokens. Preliminary image-generation experiments demonstrate the feasibility of ProxyFormer for both pixel-space and latent-space flow matching.