arXiv subjects

Karl Li

Publications and source records attributed to Karl Li.

3 recordsLinked to original sources

arXiv2026

Sparse Autoencoder Attribution Graphs Are Decided at the Margin

Attribution graphs over the features of sparse autoencoders (SAEs) are used to predict what an ablation does and are compared across training seeds. A TopK SAE keeps the largest preactivations at each token, and a feature's margin is its preactivation's distance to the cutoff. On five public TopK SAE pairs from three Pythia models, zeroing one feature moves a next layer feature across the cutoff in 0.74 of cases or more, and the crossings carry most of the response's energy. Over writes that scale a feature's code by three factors, the graph built on the features active beforehand has a median relative error of 0.247 or more. The margin graph adds the first order change to every preactivation and reranks, at the same cost, and errs by at most 0.033. At the median, the crossing features' decoded change is 1.24 times the model's own change in norm or more. Between seeds of one testbed, sparse maps between features explain 0.85 of a graph's energy on its active features, and 0.47 once the other seed's graph keeps only its own. The largest model studied is pythia-160m.

cs.LG↗

arXiv2026

Announced Breaks Separate Conformal Reliability from Frequency Calibration

Many structural breaks are announced in advance, yet adaptive conformal methods learn of them only from their own errors. A reset at the announced date serves one of two audits, conformal reliability of the next interval or frequency calibration of a closed window, and the two can disagree. For independent scores, restarting the empirical quantile at the boundary attains the minimax reliability rate, of order one over the square root of the post-break count. Restarting an adaptive recursion attains a frequency gap of order one over that count on bounded scores when its floor adds a bounded total. That gap bounds nothing about the next interval. After a large synthetic break, the restarted quantile's coverage error at the 50th forecast is 0.035, against 0.052 for the best of five adaptive baselines. The restarted recursion has the smallest frequency gap over the first 50 forecasts after every synthetic break. Real breaks are mostly smaller. The restarted quantile beats two baselines after an electricity market reform and trails all five after monetary policy decisions. Of nine announced breaks screened, only one is large enough for a detectable gain, and there every method misses nominal coverage by more than 0.1.

cs.LG↗

arXiv2022

Deep Learning Based Detection of Enlarged Perivascular Spaces on Brain MRI

BACKGROUND AND PURPOSE: Deep learning has been demonstrated effective in many neuroimaging applications. However, in many scenarios, the number of imaging sequences capturing information related to small vessel disease lesions is insufficient to support data-driven techniques. Additionally, cohort-based studies may not always have the optimal or essential imaging sequences for accurate lesion detection. Therefore, it is necessary to determine which imaging sequences are crucial for precise detection. This study introduces a novel deep learning framework to detect enlarged perivascular spaces (ePVS) and aims to find the optimal combination of MRI sequences for deep learning-based quantification. MATERIALS AND METHODS: We implemented an effective lightweight U-Net adapted for ePVS detection and comprehensively investigated different combinations of information from SWI, FLAIR, T1-weighted (T1w), and T2-weighted (T2w) MRI sequences. The training data included 21 participants, which were randomly selected from the MESA cohort. Participants had ePVS 683 lesions on average. For T1w, T2w, and FLAIR images, the MESA study collected 3D isotropic MRI scans at six different sites with Siemens scanners. Our training data included participants from all these sites and all the scanner models, and the proposed model was applied to the whole brain instead of selective regions. RESULTS: The experimental results showed that T2w MRI is the most important for accurate ePVS detection, and the incorporation of SWI, FLAIR and T1w MRI in the deep neural network had minor improvements in accuracy and resulted in the highest sensitivity and precision (sensitivity =0.82, precision =0.83). The proposed method achieved comparable accuracy at a minimal time cost compared to manual reading.

eess.IV↗