arXiv subjects

Aaditya Khanal

Publications and source records attributed to Aaditya Khanal.

4 recordsLinked to original sources

arXiv2026

Knowledge Distillation from Large Reasoning Models to Compact Student Models: A Case Study on the John O Bryan Mathematics Competition

This paper investigates knowledge distillation from a large reasoning model (DeepSeek-R1) to a compact student model (Qwen2.5-7B). Using historical problems from the John O'Bryan Mathematics Competition at Northern Kentucky University (2011-2025), we build a Chain-of-Thought (CoT) training corpus through a dual-agent framework. The dataset is used to fine-tune the student model with Low-Rank Adaptation (LoRA) on Apple Silicon hardware using the MLX framework. The base Qwen2.5-7B model achieves 64.67% accuracy on competition problems, while the DeepSeek-R1 teacher achieves 91.40%. An initial 1,000-iteration training run revealed severe overfitting, with validation loss reaching a minimum at iteration 200 before rising steadily. Based on this finding, we ran five independent training runs each limited to 200 iterations with varied random seeds to assess result stability. Across these five runs, the fine-tuned student model achieves a mean accuracy of 69.43% (std dev 0.17%) on the competition dataset, a 4.76 percentage-point improvement over the base model, and generalizes to 73.1% (std dev 0.18%) on the MATH-500 benchmark. We further study how response length affects answer quality across six reasoning levels (R1-R6): accuracy declines consistently from 69.43% at R1 (mean 220 words) to 41.9% at R6 (mean 31.2 words), with the two-person speed section most sensitive to token reduction. These results demonstrate that CoT distillation improves compact student models and that response length is a critical factor in mathematical reasoning quality.

cs.LG↗

arXiv2026

Beyond pass@1: A Reliability Science Framework for Long-Horizon LLM Agents

Existing benchmarks measure capability -- whether a model succeeds on a single attempt -- but production deployments require reliability -- consistent success across repeated attempts on tasks of varying duration. We show these properties diverge systematically as task duration grows, and that pass@1 on short tasks is structurally blind to this divergence. We introduce a reliability science framework for long-horizon LLM agents with four metrics: Reliability Decay Curve (RDC), Variance Amplification Factor (VAF), Graceful Degradation Score (GDS), and Meltdown Onset Point (MOP). We evaluate 10 models across 23,392 episodes on a 396-task benchmark spanning four duration buckets and three domains. Key findings: (1) reliability decay is domain-stratified -- SE GDS drops from 0.90 to 0.44 while document processing is nearly flat (0.74 to 0.71); (2) VAF bifurcates by capability tier -- high VAF is a capability signature, not an instability signal; (3) capability and reliability rankings diverge substantially, with multi-rank inversions at long horizons; (4) frontier models have the highest meltdown rates (up to 19%) because they attempt ambitious multi-step strategies that sometimes spiral; and (5) memory scaffolds universally hurt long-horizon performance across all 10 models. These results motivate reliability as a first-class evaluation dimension alongside capability.

cs.AI↗

arXiv2026

Severe Domain Shift in Skeleton-Based Action Recognition:A Study of Uncertainty Failure in Real-World Gym Environments

The practical deployment gap -- transitioning from controlled multi-view 3D skeleton capture to unconstrained monocular 2D pose estimation -- introduces a compound domain shift whose safety implications remain critically underexplored. We present a systematic study of this severe domain shift using a novel Gym2D dataset (style/viewpoint shift) and the UCF101 dataset (semantic shift). Our Skeleton Transformer achieves 63.2% cross-subject accuracy on NTU-120 but drops to 1.6% under zero-shot transfer to the Gym domain and 1.16% on UCF101. Critically, we demonstrate that high Out-Of-Distribution (OOD) detection AUROC does not guarantee safe selective classification. Standard uncertainty methods fail to detect this performance drop: the model remains confidently incorrect with 99.6% risk even at 50% coverage across both OOD datasets. While energy-based scoring (AUROC >= 0.91) and Mahalanobis distance provide reliable distributional detection signals, such high AUROC scores coexist with poor risk-coverage behavior when making decisions. A lightweight finetuned gating mechanism restores calibration and enables graceful abstention, substantially reducing the rate of confident wrong predictions. Our work challenges standard deployment assumptions, providing a principled safety analysis of both semantic and geometric skeleton recognition deployment.

cs.CV↗

arXiv2022

Fundamental Investigation of Reactive-Convective Transport: Implications for Long-Term Carbon dioxide (CO2) Sequestration

The density-driven convection coupled with chemical reaction is the preferred mechanism for permanently storing CO2 in saline aquifers. This study uses a 2D visual Hele-Shaw cell to evaluate and visualize the density-driven convection formed due to gravitational instabilities, also known as Rayleigh-Taylor instability. The primary goal of the experiments is to understand the various mechanisms for the mass transfer of gaseous CO2 into brine with different initial ionic concentrations and flow permeability. Moreover, the impact of CO2 flow rates, injection locations, reservoir dipping angle, and permeability heterogeneity is also investigated. We observed that the presence of salts resulted in earlier onset of convection and a larger convective finger wavelength than the case with no dissolved salts. In addition, experimental data showed a higher lateral mixing between CO2 fingers when dipping is involved. The visual investigation also revealed that the CO2 dissolution rate, measured by the rate of the convective fingers advance, depends on the type and concentration of the ions present in the brine. The CO2 dissolution for solutions with varying salt dissolved, indicated by the area of the pH-depressed region, is observed to be 0.38-0.77 times compared to when no salt is present. Although convective flow is slowed down in the presence of salts, the diffusive flux is enhanced, as observed from both qualitative and quantitative results. Moreover, the reduced formation permeability, introduced by using a flow barrier, resulted in numerous regions not being swept by the dissolved CO2, indicating an inefficient dissolution. We also investigated the effect of discrete high conductivity fractures within the flow barriers, which showed an uneven vertical sweep and enhanced flow channeling. Lastly, the parameters regarding CO2 leakage risk during storage are identified and discussed.

physics.flu-dyn↗