When Does Sparsity Mitigate the Curse of Depth in LLMs

Recent work has demonstrated the curse of depth in large language models (LLMs), where later layers contribute less to learning and representation than earlier layers. Such under-utilization is linked to the accumulated growth of variance in Pre-Layer Normalization, which can push deep blocks toward near-identity behavior. In this paper, we demonstrate that, sparsity, beyond enabling efficiency, acts as a regulator of variance propagation and thereby improves depth utilization. Our investigation covers two sources of sparsity: (i) implicit sparsity, which emerges from training and data conditions, including weight sparsity induced by weight decay and attention sparsity induced by long context inputs; and (ii) explicit sparsity, which is enforced by architectural design, including key/value-sharing sparsity in Grouped-Query Attention and expert-activation sparsity in Mixtureof-Experts. Our claim is thoroughly supported by controlled depth-scaling experiments and targeted layer effectiveness interventions. Across settings, we observe a consistent relationship: sparsity improves layer utilization by reducing output variance and promoting functional differentiation. We eventually distill our findings into a practical rule-of-thumb recipe for training deptheffective LLMs, yielding a notable 4.6% accuracy improvement on downstream tasks. Our results reveal sparsity, arising naturally from standard design choices, as a key yet previously overlooked mechanism for effective depth scaling in LLMs. Code is available at https://github.com/pUmpKin-Co/SparsityAndCoD.

The Agentic Researcher: A Practical Guide to AI-Assisted Research in Mathematics and Machine Learning

AI tools and agents are reshaping how researchers work, from proving theorems to training neural networks. Yet for many, it remains unclear how these tools fit into everyday research practice. This paper is a practical guide to AI-assisted research in mathematics and machine learning: We discuss how researchers can use modern AI systems productively, where these systems help most, and what kinds of guardrails are needed to use them responsibly. It is organized into three parts: (I) a five-level taxonomy of AI integration, (II) an open-source framework that, through a set of methodological rules formulated as agent prompts, turns CLI coding agents (e.g., Claude Code, Codex CLI, OpenCode) into autonomous research assistants, and (III) case studies from deep learning and mathematics. The framework runs inside a sandboxed container, works with any frontier LLM through existing CLI agents, is simple enough to install and use within minutes, and scales from personal-laptop prototyping to multi-node, multi-GPU experimentation across compute clusters. In practice, our longest autonomous session ran for over 20 hours, dispatching independent experiments across multiple nodes without human intervention. We stress that our framework is not intended to replace the researcher in the loop, but to augment them. Our code is publicly available at https://github.com/ZIB-IOL/The-Agentic-Researcher.

ECHOSAT: Estimating Canopy Height Over Space And Time

Forest monitoring is critical for climate change mitigation. However, existing global tree height maps provide only static snapshots and do not capture temporal forest dynamics, which are essential for accurate carbon accounting. We introduce ECHOSAT, a global and temporally consistent tree height map at 10 m resolution spanning multiple years. To this end, we resort to multi-sensor satellite data to train a specialized vision transformer model, which performs pixel-level temporal regression. A self-supervised growth loss regularizes the predictions to follow growth curves that are in line with natural tree development, including gradual height increases over time, but also abrupt declines due to forest loss events such as fires. Our experimental evaluation shows that our model improves state-of-the-art accuracies in the context of single-year predictions. We also provide the first global-scale height map that accurately quantifies tree growth and disturbances over time. We expect ECHOSAT to advance global efforts in carbon monitoring and disturbance assessment. The maps can be accessed at https://github.com/ai4forest/echosat.

Investigating the Influence of Spatial Ability in Augmented Reality-assisted Robot Programming

Augmented Reality (AR) offers promising opportunities to enhance learning, but its mechanisms and effects are not yet fully understood. As learning becomes increasingly personalized, considering individual learner characteristics becomes more important. This study investigates the moderating effect of spatial ability on learning experience with AR in the context of robot programming. A between-subjects experiment ($N=71$) compared conventional robot programming to an AR-assisted approach using a head-mounted display. Participants' spatial ability was assessed using the Mental Rotation Test. The learning experience was measured through the System Usability Scale (SUS) and cognitive load. The results indicate that AR support does not significantly improve the learning experience compared to the conventional approach. However, AR appears to have a compensatory effect on the influence of spatial ability. In the control group, spatial ability was significantly positively associated with SUS scores and negatively associated with extraneous cognitive load, indicating that higher spatial ability predicts a better learning experience. In the AR condition, these relationships were not observable, suggesting that AR mitigated the disadvantage typically experienced by learners with lower spatial abilities. These findings suggest that AR can serve a compensatory function by reducing the influence of learner characteristics. Future research should further explore this compensatory role of AR to guide the design of personalized learning environments that address diverse learner needs and reduce barriers for learners with varying cognitive profiles.

From Associations to Activations: Comparing Behavioral and Hidden-State Semantic Geometry in LLMs

We investigate the extent to which an LLM's hidden-state geometry can be recovered from its behavior in psycholinguistic experiments. Across eight instruction-tuned transformer models, we run two experimental paradigms -- similarity-based forced choice and free association -- over a shared 5,000-word vocabulary, collecting 17.5M+ trials to build behavior-based similarity matrices. Using representational similarity analysis, we compare behavioral geometries to layerwise hidden-state similarity and benchmark against FastText, BERT, and cross-model consensus. We find that forced-choice behavior aligns substantially more with hidden-state geometry than free association. In a held-out-words regression, behavioral similarity (especially forced choice) predicts unseen hidden-state similarities beyond lexical baselines and cross-model consensus, indicating that behavior-only measurements retain recoverable information about internal semantic geometry. Finally, we discuss implications for the ability of behavioral tasks to uncover hidden cognitive states.

Within-Model vs Between-Prompt Variability in Large Language Models for Creative Tasks

How much of LLM output variance is explained by prompts versus model choice versus stochasticity through sampling? We answer this by evaluating 12 LLMs on 10 creativity prompts with 100 samples each (N = 12,000). For output quality (originality), prompts explain 36.43% of variance, comparable to model choice (40.94%). But for output quantity (fluency), model choice (51.25%) and within-LLM variance (33.70%) dominate, with prompts explaining only 4.22%. Prompts are powerful levers for steering output quality, but given the substantial within-LLM variance (10-34%), single-sample evaluations risk conflating sampling noise with genuine prompt or model effects.

SparseSwaps: Tractable LLM Pruning Mask Refinement at Scale

The resource requirements of Neural Networks can be significantly reduced through pruning -- the removal of seemingly less important parameters. However, with the rise of Large Language Models (LLMs), full retraining to recover pruning-induced performance degradation is often prohibitive and classical approaches such as global magnitude pruning are suboptimal on Transformer architectures. State-of-the-art methods hence solve a layer-wise mask selection problem, the problem of finding a pruning mask which minimizes the per-layer pruning error on a small set of calibration data. Exactly solving this problem to optimality using Integer Programming (IP) solvers is computationally infeasible due to its combinatorial nature and the size of the search space, and existing approaches therefore rely on approximations or heuristics. In this work, we demonstrate that the mask selection problem can be made drastically more tractable at LLM scale. To that end, we decouple the rows by enforcing equal sparsity levels per row. This allows us to derive optimal 1-swaps (exchanging one kept and one pruned weight) that can be computed efficiently using the Gram matrix of the calibration data. Using these observations, we propose a tractable and simple 1-swap algorithm that warm starts from any pruning mask, runs efficiently on GPUs at LLM scale, and is essentially hyperparameter-free. We demonstrate that our approach reduces per-layer pruning error by up to 60% over Wanda (Sun et al., 2023) and consistently improves perplexity and zero-shot accuracy across state-of-the-art GPT architectures.

A Free Lunch in LLM Compression: Revisiting Retraining after Pruning

While Neural Network pruning typically requires retraining the model to recover pruning-induced performance degradation, state-of-the-art Large Language Models (LLMs) pruning methods instead solve a layer-wise mask selection and reconstruction problem on a small set of calibration data to avoid full retraining, as it is considered computationally infeasible for LLMs. Reconstructing single matrices in isolation has favorable properties, such as convexity of the objective and significantly reduced memory requirements compared to full retraining. In practice, however, reconstruction is often implemented at coarser granularities, e.g., reconstructing a whole transformer block against its dense activations instead of a single matrix. In this work, we study the key design choices when reconstructing or retraining the remaining weights after pruning. We conduct an extensive computational study on state-of-the-art GPT architectures, and report several surprising findings that challenge common intuitions about retraining after pruning. In particular, we observe a free lunch scenario: reconstructing attention and MLP components separately within each transformer block is nearly the most resource-efficient yet achieves the best perplexity. Most importantly, this Pareto-optimal setup achieves better performance than full retraining, despite requiring only a fraction of the memory. Furthermore, we demonstrate that simple and efficient pruning criteria such as Wanda can outperform much more complex approaches when the reconstruction step is properly executed, highlighting its importance. Our findings challenge the narrative that retraining should be avoided at all costs and provide important insights into post-pruning performance recovery for LLMs.

Scalable DC Optimization via Adaptive Frank-Wolfe Algorithms

We consider the problem of minimizing a difference of (smooth) convex functions over a compact convex feasible region $P$, i.e., $\min_{x \in P} f(x) - g(x)$, with smooth $f$ and Lipschitz continuous $g$. This computational study builds upon and complements the framework of Maskan et al. [2025] by integrating advanced Frank-Wolfe variants to reduce computational overhead. We empirically show that constrained DC problems can be efficiently solved using a combination of the Blended Pairwise Conditional Gradients (BPCG) algorithm [Tsuji et al., 2022] with warm-starting and the adaptive error bound from Maskan et al. [2025]. The result is a highly efficient and scalable projection-free algorithm for constrained DC optimization.

Neural Concept Verifier: Scaling Prover-Verifier Games via Concept Encodings

While Prover-Verifier Games (PVGs) offer a promising path toward verifiability in nonlinear classification models, they have not yet been applied to complex inputs such as high-dimensional images. Conversely, Concept Bottleneck Models (CBMs) effectively translate such data into interpretable concepts but are limited by their reliance on low-capacity linear predictors. In this work, we introduce the Neural Concept Verifier (NCV), a unified framework combining PVGs with concept encodings for interpretable, nonlinear classification in high-dimensional settings. NCV achieves this by utilizing recent minimally supervised concept discovery models to extract structured concept encodings from raw inputs. A prover then selects a subset of these encodings, which a verifier -- implemented as a nonlinear predictor -- uses exclusively for decision-making. Our evaluations show that NCV outperforms CBM and pixel-based PVG classifier baselines on high-dimensional, logically complex datasets and also helps mitigate shortcut behavior. Overall, we demonstrate NCV as a promising step toward performative, verifiable AI.