Computer Science (arXiv)

Pipeline parallelism (PP) is widely used to scale large language model (LLM) training, but its efficiency is often limited by stage imbalance and pipeline bubbles. Meanwhile, cross-layer KV sharing has primarily been studied as a mechanism for reducing KV-cache costs during inference, without examining how KV reuse reshapes pipeline workloads. We present \textbf{KV-Pipe}, a stage-aware KV-sharing mechanism that turns KV reuse into a pipeline-balancing control knob. KV-Pipe starts from the tail stage, converts selected attention layers to cross-layer KV sharing in a tail-first order, and iteratively retargets the current bottleneck to drive the FLOPs Imbalance Ratio (FIR) toward $1$. The procedure is performed offline and requires only a pipeline partition and per-layer FLOPs estimates, introducing negligible runtime overhead and requiring no online tuning. Across multiple pipeline-parallel configurations, KV-Pipe consistently improves utilization and throughput, achieving up to \textbf{9.2\%} higher training MFU and up to a \textbf{9.8\%} reduction in iteration time, with larger gains at higher pipeline-parallel degrees where stage imbalance is amplified. Furthermore, the same KV-sharing mechanism provides an inference-side benefit by reducing KV-cache growth and redundant KV projection work, resulting in higher decoding throughput for long-context workloads. These results identify KV layout as a system--architecture degree of freedom for jointly improving pipeline-parallel training efficiency and long-context inference.

Early multi-agent LLM systems often used role-separated teams, yet scaling agent count yields diminishing returns on repository-level coding tasks. Recent alternatives treat agents as passive tools (subagents), yet this removes the benefits of agent interaction entirely. We study whether a subagent paradigm can support a middle ground: minimal agentic cooperation without the overhead of large multi-agent teams. We introduce Adversarial Review (AR), a minimal cooperative code-review protocol in which a main coding agent works with a reviewer and a critic agent. The reviewer evaluates code, while the critic audits the review through structured disagreement before the main agent edits. On LiveCodeBench, AR achieves the highest pass rate among tested methods, outperforming a five-agent baseline while using only three agents. On SWE-PRBench, naive AR exposes a false-consensus failure mode, where agents converge on agreement without sufficient evidence, but a single prompt iteration that adds disagreement explicitly achieves the highest F1 among tested methods. On SWE-bench Verified, AR also shows improvements over the baselines on repository-level coding tasks. Together, AR demonstrates that cooperative code review does not require many agents or complex communication structures: it requires that disagreement be minimal, structured, and evidence-grounded.

Modern encoder-decoder systems can produce fluent text even when their input contains no recoverable message. We study this failure in ASR and NMT through the models' reserved null tokens, asking whether the score for ending generation already carries a usable abstention signal. Across speech recognizers and translation models, we audit native null-token scores and scalar logit shifts. In Whisper, we additionally probe decoder states and compare supervised row edits with conventional external gates. The evaluated models often expose a useful abstention signal, but stock decoding does not reliably act on it. Raising the null-token score can sharply suppress fabrication, but aggressive intervention also deletes valid speech or shortens legitimate translations. These findings turn the null token into a diagnostic lens on hallucination and motivate evaluating abstention methods by both suppression and deletion costs, rather than by hallucination reduction alone.

Stateful language agents assume a rejected branch can be taken back by clearing it from the application transcript. We show this breaks when the serving session retains key/value (KV) state across the logical abort: the model can continue attending to content the application believes it discarded. We formalize the missing guarantee as rollback consistency: a complete abort must restore the state the model attends, not just the transcript. The key failure is cross-layer: a correct logical rollback need not compose with retained inference state, and the gap can remain invisible to the application. To isolate cache effects from text effects, we introduce a same-token/different-cache audit that holds decision-step tokens identical while varying only whether the cached prefix is stale or rebuilt from committed state. Across seven open-weight families (3.8B-36B), retained KV alone flips a typed protected effect in 25 of 63 audited cells, while attacker tokens are absent from the served request in all 63; rebuilding the cache closes every cell. The channel reproduces in an end-to-end session application, on the default Hugging Face Transformers cache-reuse path, and under LangGraph time-travel, where verified logical rollback can still leave attended KV stale. Susceptibility varies across models, but the underlying attended-state integrity violation is structural. We rule out position and length confounds, generalize across protected effects, policy structures, and a cache-isolated Mixture-of-Experts model, and show that transaction-local cache restoration closes the channel without requiring a global cache flush. All headline results are deterministic and reproducible from released artifacts.

Action chunking --- the practice of predicting a sequence of actions and executing a prefix open-loop --- has emerged as a key enabler of recent progress in imitation learning for robotic manipulation. However, executing long open-loop prefixes reduces reactivity, limiting policies' ability to correct for errors. Further, the mechanisms underlying these performance benefits remain poorly understood: prior works cite mitigating compounding errors, absorbing inference latency, or smoothing motions, but provide limited controlled evidence or guidance for preserving reactivity. In this work, we argue that long open-loop execution primarily helps short-context policies imitate "non-Markovian demonstrations". Across four simulation and two real-world tasks, we show that expert non-Markovianity strongly shapes the relationship between task success and open-loop execution horizon. Further, we investigate the impact of compounding errors --- the prevailing explanation for long open-loop execution in prior work --- and find that while they matter, expert non-Markovianity has a much stronger impact in our experimental setting. Finally, we show that when policies are provided with a sufficiently long context, open-loop execution is no longer beneficial and the most reactive, closed-loop policies perform best. While imitation learning has seen great success using long open-loop execution, our findings motivate long-context, reactive policies as a more principled and performant paradigm.

In the theory of error correcting codes, list-decoding refers to the following problem. Given a code $C \subseteq Σ^N$ and a received word $y \in Σ^N$, find all codewords $c \in C$ so that $δ(c,y) \leq ρ$, where $δ$ is relative Hamming distance and $ρ\in (0,1)$. Codes that approach the optimal trade-off between the rate $R := \log_{|Σ|}(|C|) / N$ and the list-decoding radius $ρ$ are said to achieve capacity.By now, there are constructions of capacity-achieving list-decodable codes with fast near-linear-time list-decoding algorithms, but most existing work has not considered space complexity. In a recent line of work, Cook and Moshkovitz (2024, 2025, 2026) initiated the study of low-space deterministic algorithms for error correcting codes. In particular, in their 2026 paper, they gave a construction of list-decodable codes with deterministic near-linear-time and sublinear space list-decoding algorithms. However, these codes were far from achieving capacity. In this paper, we present list-decodable codes approaching capacity with deterministic time- and space-efficient list-decoding algorithms. More precisely, for any $R \in (0,1)$ and any arbitrarily small constant $τ> 0$, we present a family of codes $C\subseteq Σ^N$ with rate $R$ that are deterministically list-decodable up to radius $ρ= 1 - R - τ$, in time $N^{1 + τ}$ and space $N^τ$ with constant output list size and constant alphabet size. Our results can be extended to capacity-achieving list-recoverable codes.

Jointly fine-tuning an LLM on meeting-summarisation corpora of widely varying size raises a question that prior work leaves confounded: when a domain-balanced training mixture helps, is the gain due to the distribution of tokens across domains, or merely to the volume of data seen? We disentangle these factors by constructing balanced and natural (native-proportional) token mixtures at matched token budgets (2-32M) over five English meeting corpora, fine-tuning Mistral-7B with QLoRA, and evaluating per domain. Balancing redistributes quality, improving the data-scarce minority domains at a low cost to the data-rich ones. The trade favours balancing whenever the minority domains matter: their share under proportional allocation is fixed at 1-2% regardless of budget, so matching balanced quality on those domains requires far more total data. We further find that pruning low-value transcript lines removes ~15% of tokens from the conversational corpora at no measurable cost, and that balancing by tokens is not the same as balancing by examples. A two-annotator study of 741 judge-labelled facts validates our fact-level evaluation. Together these results give practitioners a basis for deciding when to balance an imbalanced multi-domain mixture, and on what unit.

Different shape representations facilitate different computations. Surface representations, in particular meshes, are often used for modeling, whereas volume representations are useful for spatial queries such as intersection or containment. Optimizing a surface representation based on a volumetric properties by gradient descent requires the derivatives of the volume relative to its bounding surface. We derive this gradient for winding numbers and show that it can be efficiently computed for volumetric values sampled on a regular grid (voxel representation) and surface parameters based on vertex sets (triangle meshes). This enables an efficient solution for a variety of optimization problems. We demonstrate the practical use of this approach at the examples of deforming meshes to resolve intersections, being manufacturable by cutting with a bandsaw from three directions, and creating shapes that are close to tiling 3D space.

Implicit surface representations have regained popularity because of their use in machine learning. A common component in optimization is regularization, penalizing the deviation of the surface from its original shape. The popular as-rigid-aspossible (ARAP) energy strikes a good compromise between realistic deformation behavior and efficient computation, at least for piecewise linear meshes. We develop an approach for computing the ARAP energy of a deformation function based on point sampling of the surface. The implicit representation is exploited to provide differentials in each sample. The evaluation is efficient and exact in each sample (up to numerical precision). We demonstrate the general applicability of the method to neural shape processing in several applications and contrast its properties with alternatives from the literature.

As state-of-the-art machine translation models saturate standard benchmarks, the field needs more challenging evaluations to distinguish between models of varying quality. We propose augmenting existing benchmarks to increase translation difficulty by combining adversarial optimization with a differentiable translation difficulty estimator. Our Adversarial Translation Optimization (ATO) uses gradients from a combined difficulty and fluency objective to iteratively replace tokens. Because each step branches over candidate substitutions at every position, optimization becomes a tree search problem, which we address with Beam Search. ATO offers a gradient-based alternative to LLM-based dataset creation without LLM prompting, expensive human curation, or task-specific model training. Our ATO-modified benchmark lowers average translation quality (xCOMET) from 0.93 to 0.82, compared to 0.88 for paraphrasing and 0.86 for a zero-shot baseline. Human evaluation shows the modified texts are somewhat less natural than the baselines but remain reasonably grammatical and plausible while being substantially harder to translate. We release two datasets of 350 English texts each, generated by our methods, as well as the code.

We present PLSQLBench, to our knowledge the first benchmark for evaluating whether LLMs can write executable PL/SQL programs, with correctness measured through execution-based tests. Existing LLM evaluations largely target general-purpose code generation or declarative text-to-SQL, leaving procedural database programming underexplored. PLSQLBench contains 2,865 instances: 2,594 single-turn tasks and 271 multi-turn conversations spanning 978 turns. The benchmark combines complex schema-grounded tasks over enterprise-style Spider 2 databases, simpler schema-grounded tasks derived from Spider, and MBPP-derived procedural problems, covering varying levels of database grounding and procedural complexity. Experiments with eight LLMs reveal recurring difficulties in schema grounding, PL/SQL dialect fidelity, procedural control flow, exception handling, and cross-turn consistency. Tool-augmented LLM agents improve performance on several schema-grounded evaluations, although substantial gaps remain. These results highlight procedural database programming capabilities not directly assessed by conventional code generation or text-to-SQL benchmarks. Our code is available at https://github.com/oracle-samples/plsqlbench.

Foundation GUI agents can automate complex digital tasks, but deployment is hindered by scarce and biased training data, ambiguous prompts, and unreliable execution. Routine workflows rely on user-specific tools and tacit conventions, so unstated instructions can produce arbitrary variations across runs. We present UI-Mate, a foundation GUI agent that integrates an environment-grounded training stack with in-context demonstration learning. UI-Mate makes three contributions: A Scalable Environment-Grounded Training Stack: A closed-loop data engine automates task generation, environment construction, rollout, filtering, capability balancing, SFT, and online RL across massively parallel environments via unified task-verifier bundles. In-Context Demonstration Learning: A mechanism that transforms multimodal demonstrations into flexible subtask-level workflows, follows relevant demonstrated steps, and re-plans from the live interface. OSWorkerBench Benchmark and Insights: A benchmark of 100 long-horizon office tasks across 41 applications that supports instruction-only and demonstration-guided evaluation. Its demonstration resources separate a 33-task self-demo setting, built from successful strong-agent rollouts of the same targets, from a 45-task variant-demo setting, built from human recordings of related but non-identical tasks. Experiments show that UI-Mate-27B sets a new open-weight state of the art on general computer-use benchmarks, scoring 77.0% on OSWorld-Verified and 66.2% on WindowsAgentArena. On OSWorkerBench, it reaches 41.0% strict success and 76.9% progress, outperforming its Qwen3.6-27B base by 17.7 and 24.5 points. On the 33-task self-demo subset, one demonstration raises strict success from 17.2% to 35.4% and progress from 67.9% to 81.1%, substantially improving long-horizon reliability. Project page: https://ui-mate.github.io.

Pedestrian path prediction is crucial for enhancing the safety of autonomous vehicles and advanced driver-assistance systems. Previous studies explored different learning-task formulations for pedestrian path prediction and compared these formulations using shallow neural networks, but did not extend this analysis to more complex deep-learning models. This paper adapts the Spatial-Temporal Graph Attention Network (STGAT) to a unified pedestrian path prediction framework and introduces state and action definitions specific to STGAT. The resulting formulations support deterministic and stochastic policies, one-time and sequential decision-making, and reinforcement-learning algorithms including REINFORCE and proximal policy optimization. The proposed learning-task formulations improve prediction performance across the selected benchmark datasets compared with the standard supervised-learning formulation. These results demonstrate that reformulating the decision process and training objective can improve an advanced pedestrian trajectory prediction architecture and may provide a path toward improving other graph-based prediction models.

Continuous graph-query engines must process edge updates while making current query results available to concurrent consumers. Existing dynamic constant-delay enumeration methods provide strong per-answer delay guarantees, but are commonly formulated as a maintenance-then-enumeration process. Conversely, multicore graph-stream engines emphasize update throughput and match discovery without a query-level snapshot guarantee for concurrent full-result enumeration. We present TrieGS, a shared-memory engine that maintains a query-specific CDE state over a streaming graph. Logically, TrieGS uses the classical free-connex witness-subtree enumerator; dynamically, it maintains multiplicity payloads using exact signed deltas; physically, RDF terms are dictionary-encoded and the required access structures are realized as versioned LFNT relations (Lock-Free Nested Trie). The new systems problem is not relation-level snapshotting alone: an enumeration job must observe one consistent state across all interdependent base relations, projections, views, and indexes. TrieGS therefore publishes an atomic root vector only after an update epoch has fully propagated. Enumeration threads pin one published root vector and traverse immutable versions while later updates continue.

Global workspace theory explains conscious access as the broadcasting of selected information to the rest of the network, but it lacks a formal criterion for identifying the mechanism that enables this access. We propose that a global workspace is a mediator, namely, a subnetwork that receives activity from distributed systems, transforms it through internal modes, and returns differentiated effects to the broader network. We formalize this claim as the Global Mediation Workspace (GMW), a control-theoretic formulation in which a candidate subnetwork is treated as an open system embedded in the remainder of the network. In this framework, reachability characterizes how the remainder can drive the candidate, observability characterizes how candidate states affect the remainder, and a boundary Hankel operator identifies the internal modes linking the two directions. The resulting signature quantifies mediation capacity, input-output alignment, effective dimensionality, and routed source-target breadth, each of which characterizes different components of global workspace. In synthetic benchmarks, we tested whether the signature can distinguish a planted differentiated mediator from dense hubs, one-sided receivers or broadcasters, and a split read/write aggregate with no common internal route. A nonlinear extension characterizes mediation through trajectory-conditioned differential operators, finite-amplitude response profiles, and state-dependent coalitions. As a preliminary application, we estimated the signature from ECoG recordings in four macaques under ketamine anesthesia. We found that input-output alignment was reduced during unconsciousness whereas potential capacity was increased. The GMW thus provides a formal and testable criterion for locating candidate global workspaces in neural recordings and for asking which aspects of mediation is crucial for conscious access.

Cloud-side inference gives imitation-learning policies access to greater computational resources, but communication and computation delays can degrade control performance. To compensate for these delays, we propose RAPAC-DP, a response-aligned pending-action compensation framework designed for both diffusion- and flow-based action generators. RAPAC-DP encodes the actions already scheduled for execution before the cloud response arrives into a pending-action sequence that serves as the conditioning input to a parameter-efficient compensation pathway. When delay effects are negligible, bypassing this pathway exactly recovers the frozen base policy. For training, RAPAC-DP constructs delay-conditioned samples from delay-free demonstrations, requiring neither explicit system dynamics nor additional delayed demonstrations. At the largest fixed delay tested on Kinetix, RAPAC-DP retained 81.4% of its overall delay-free performance. At the largest fixed delay tested on each RoboMimic task, it achieved a mean success rate of 0.633 across the three tasks. These results demonstrate the effectiveness of pending-action compensation for cloud-deployed imitation-learning policies.

We show that the natural-gradient stationary condition of variational inference has an edge-local form on a Forney-style factor graph. We start from the Bethe free energy and constrain a selected edge marginal to an exponential family. At a stationary point, the natural parameter of that edge equals the sum of two projected messages, one from each incident factor. Each projected message is the natural-gradient projection of the exact belief-propagation log-message at the current receiving marginal, or equivalently, the gradient of its expectation in the so-called mean coordinates. We call the resulting scheme natural-gradient message passing (NGMP). The rule is local; each edge may carry its own exponential family, and the message a factor sends depends on the marginal that receives it. Compared with variational message passing, NGMP keeps the part of the exact message that the receiving family can represent instead of averaging the factor under the neighboring beliefs. The two coincide when the uncertainty on the edges entering a non-conjugate factor vanishes, and NGMP is more accurate when that uncertainty persists, for example, along a partially observed latent chain or when parameters are filtered through successive data batches. Experiments on Poisson smoothing, heteroskedastic regression, and hourly ETTh forecasting confirm this and show that the gain appears mainly in uncertainty calibration.

The Lightning Network (LN) utilizes gossip to share network topology, channel announcements and updates, and node announcements among its local constituents. Yet, our measurements show that this flooding-based gossip reconciliation is fundamentally inefficient. We propose, instead, to use set reconciliation protocols for sharing this information, and we systematically evaluate existing approaches under realistic network conditions. We further propose ADAPTIVEIBLT, a novel adaptive IBLT (Invertible Bloom Lookup Table) protocol with a partial-decoding enhancement. By simulating reconciliation in Core-Lightning and evaluating real gossip snapshots, we demonstrate the practical benefits of reconciliation in scaling gossip reconciliation from hours down to a few minutes.

Intelligent systems are increasingly deployed in safety-critical settings with black-box controllers, including neural networks. The properties and behaviors of these end-to-end systems can be studied with abstraction-based methods that replace them with simpler finite models. Constructing such abstractions requires balancing the soundness of over-approximating the dynamical system against conservatism, which manifests as spurious or excessive nondeterministic behaviors. Bi-simulation theory provides principled metrics for characterizing these relationships, but does not prescribe how to construct sound abstractions with minimal conservatism. We fill this gap with a smooth simulation surrogate ($S^3$) --- a differentiable objective that approximates the reverse simulation metric used to quantify conservatism. Combined with Taylor model-based reachability, $S^3$ enables gradient-based optimization of abstraction parameters while preserving soundness by construction. We evaluate this optimization pipeline on three case studies. Our results show that $S^3$ is strongly correlated with the reverse simulation metric, is computationally faster, and serves as an effective objective for reducing abstraction conservatism.

Retrieval-Augmented Generation over knowledge graphs (Graph-RAG) has emerged as a powerful paradigm for grounding large language models in domain-specific corpora. However, existing systems face persistent limitations: (1) static chunking fragments long documents, losing cross-section semantic connections; (2) ingestion pipelines do not scale adaptively; and (3) multi-domain deployments require either a monolithic knowledge base that dilutes retrieval precision or manual user routing. We present Noesis, a decoupled Graph-RAG architecture addressing these limitations through four algorithms: (a) Bidirectional Graph Traversal with a Graph-Feedback Context Resolver simulating human reading with degrading memory; (b) an AIMD Concurrency Controller adapted from TCP congestion control, achieving 23x speedup with zero OOM events; (c) Moesis, domain-aware selective quantization for MoE models achieving 6.3x speedup on 12 GB consumer GPUs; and (d) Mesh, cross-KB semantic routing with runtime structural discovery enabling small on-premises models to perform multi-hop cross-domain reasoning. On HotpotQA (1,000 questions), Noesis achieves 59.5 EM / 74.7 F1, surpassing GraphRAG by +27.8 EM while using a 35B on-premises model for graph construction rather than GPT-4o. Source text verification on a 193-page document confirms 90% precision on long-range causal edges inaccessible to chunk-independent extraction.

Large-scale pre-training has made robot policy fine-tuning increasingly data-efficient, but this progress has largely been driven by datasets and embodiments built around simple parallel-jaw grippers. Dexterous, multi-fingered hands remain comparatively data-starved because real teleoperation is costly to scale, while human hand video is off-embodiment and requires lossy pose estimation and retargeting. We introduce Simulation Pre-training for Dexterity (SPD), a pre-training framework for dexterous manipulation that uses data entirely collected in simulation. In SPD, humans manipulate virtual objects inside a VR headset, enabling on-embodiment trajectories and robot-free collection. With the help of five operators, we collect 75 hours of multi-task dexterous manipulation over one week, and use it to pre-train a causal transformer on a sequence modeling objective. We study the benefits of simulation pre-training on real-world tasks by fine-tuning on 1-2 hours of physical demonstrations on a 56-DoF bimanual dexterous setup. We find that our approach outperforms training behavior cloning policies from scratch, showing that simulation teleoperation is a viable pre-training source for real-world dexterous manipulation. We perform ablation studies, measuring the benefits of history conditioning and short action chunks for reactive control.

Lung cancer remains the leading cause of cancer-related mortality worldwide, while histopathological diagnosis is often affected by inter-observer variability and the substantial workload associated with manual slide examination. Although deep learning has shown considerable potential in computational pathology, comprehensive benchmarks that integrate tissue classification and region segmentation within a unified analytical framework remain limited. This study presents a two-stage deep learning framework for multi-class tissue classification and pixel-level histopathological region segmentation, accompanied by a systematic comparison of state-of-the-art architectures at each stage. For tissue classification, six models, a custom convolutional neural network, VGG16, DenseNet, MobileNetV3, a custom Vision Transformer, and YOLO11, are evaluated on a combined dataset of 39,000 images derived from LC25000 and LungHist700. The models distinguish between adenocarcinoma, squamous cell carcinoma, and normal lung tissue. YOLO11 achieves the best classification performance, with an accuracy of 98.38%, a five-fold cross-validation accuracy of 98.21 +/- 0.35%, and a macro F1-score of 0.98. For region segmentation, U-Net, ResNet-encoder U-Net, DeepLabV3+, and YOLO11-seg are evaluated using the GlaS gland segmentation benchmark. DeepLabV3+ obtains the highest Intersection over Union of 0.80 and a Dice score of 0.89, while YOLO11-seg achieves a comparable Intersection over Union of 0.79 using approximately 14x fewer parameters. The best-performing classification and segmentation models are subsequently integrated into an end-to-end framework, providing an accurate, computationally efficient, and reproducible baseline for automated histopathological image analysis.

Large Language and Vision-Language Models are increasingly deployed through inference pipelines that include prompt wrappers (e.g., templates and post-processing scripts) and configuration metadata (e.g., JSON/YAML files) that together shape model outputs. While model weights and binaries are routinely verified, these textual deployment artifacts remain weakly protected despite directly influencing runtime behavior. We show that a malicious developer can pair a benign-looking wrapper with crafted metadata to deterministically alter post-generation behavior without modifying model weights, training data, or inference backend. We study this behavior through a controlled conjunctive-gate implementation, where activation depends on both an embedded wrapper marker and cryptographically bound metadata. We evaluate the attack across fifteen open- and closed-source LLM/VLM deployments, and assess prompt and system level defenses including static metadata inspection, wrapper scanners, PromptShield, and SigStore-based artifact signing. To mitigate this risk, we introduce TIF-BAH, a lightweight middleware defense that verifies wrapper integrity and records behavioral attestations during inference. Our results reveal that wrapper-metadata interactions form an under-protected execution layer in modern AI deployments, exposing a deployment-time behavioral risk that is not captured by model-weight or prompt-level defenses. Code is available at https://github.com/N-H-Arif/llm_temp.

Expressive text-to-speech (TTS) systems that use explicit conditioning labels provide direct and interpretable control over expressive attributes, in contrast to reference-based or prompting-based approaches, but require labeled data. Obtaining these labels at scale is costly and time-consuming, yet no prior semi-supervised framework addresses this specific bottleneck. Existing semi-supervised TTS methods instead target scarcity of paired speech-text data or transcriptions. To address the scarcity of expressive labels, we propose an Iterative Self-Learning (ISL) framework for expressive TTS, built on Invert-Classify, a classifier-free method that recovers discrete expressive labels by inverting a frozen generative model. The framework iteratively pseudo-labels unlabeled speech using the current model, retrains on the combined labeled and pseudo-labeled data, and repeats, progressively refining label quality and synthesis. We validate on two expressive tasks, word-level prominence and utterance-level emotion, across multiple low-resource data splits. We find that iterative refinement can improve pseudo-label accuracy over single-pass baselines. Furthermore, we observe that these improvements in pseudo-labeling of expressivity translate to gains in expressive label adherence and synthesis quality, confirmed by objective metrics and human listening tests. In the most data-scarce conditions, ISL-trained models outperform single-pass pseudo-labeling and further approach fully supervised performance, demonstrating that gradient-based ISL is an effective solution to expressive label scarcity in low-resource TTS.

Background: Planning multi-study analyses requires identifying cohorts with the relevant participants, phenotypes, and data modalities. This process commonly relies on prior knowledge, cohort catalogues, and manual literature searches. We developed a complementary question-driven framework that searches relevant scientific literature and extracts explicit cohort names. Methods: The framework first generates multiple PubMed queries from configurable vocabularies and templates and retrieves the resulting scientific literature automatically through the PubMed API. A large language model then screens the retrieved titles and abstracts and extracts explicit cohort names using a prompt tailored to the research question. The extracted names are deduplicated with human review. Configurable code, prompts, and example outputs are available at https://gitlab.rz.uni-frankfurt.de/cap_molgenlab/literature-cohort-discovery. Evaluation: As a use case, we applied the framework to youth aggression genetics. From 5,400 generated PubMed queries, the framework retrieved 5,254 unique records and identified 188 candidate cohorts. Manual screening using predefined criteria, including participant age and genetic-data availability, retained 44 eligible cohorts. Automated LLM-based name extraction was within the agreement range of human annotators. We also searched four established cohort catalogues using the same research question. Their combined results contained 27 of the 44 eligible cohorts, while 17 were not returned by any cohort catalogue search. Conclusion: The framework converts research-question-specific vocabulary into screenable cohort inventories via a large, automated literature search. It can be adapted across populations, phenotypes, data modalities, and study designs, and provides a literature-based complement to curated cohort catalogues.