Tuning tensor program generation involves navigating a vast search space to find optimal program transformations and measurements for a program on target hardware. The complexity of this process is further amplified by the exponential combinations of transformations, especially in heterogeneous environments. This research addresses these challenges by introducing a novel approach that learns the joint neural network and hardware features space, facilitating knowledge transfer to new, unseen target hardware. A comprehensive analysis is conducted on the existing state-of-the-art dataset, TenSet, including a thorough examination of test split strategies and the proposal of methodologies for dataset pruning. Leveraging an attention-inspired technique, we tailor the tuning of tensor programs to embed both neural network and hardware-specific features. Notably, our approach substantially reduces the dataset size by up to 53% compared to the baseline without compromising Pairwise Comparison Accuracy (PCA). Furthermore, our proposed methodology demonstrates competitive or improved mean inference times with only 25%40% of the baseline tuning time across various networks and target hardware. The attention-based tuner can effectively utilize schedules learned from previous hardware program measurements to optimize tensor program tuning on previously unseen hardware, achieving a top-5 accuracy exceeding 90%. This research introduces a significant advancement in auto-tuning tensor program generation, addressing the complexities associated with heterogeneous environments and showcasing promising results regarding efficiency and accuracy.

Purpose Isocitrate dehydrogenase (IDH) mutation has become one of the most important prognostic biomarkers in glioma management, indicating better treatment response and prognosis. IDH mutations confer neomorphic activity leading to the conversion of alpha-ketoglutarate (α-KG) to 2-hydroxyglutarate (2HG). The purpose of this study was to investigate the clinical potential of proton MR spectroscopy (1H-MRS) in identifying IDH-mutant gliomas by detecting characteristic resonances of 2HG and its complex interplay with other clinically relevant metabolites. Materials and Methods Thirty-two patients with suspected infiltrative glioma underwent single-voxel (SVS, n=17) and/or single-slice multivoxel (1H-MRSI, n=15) proton MR spectroscopy (1H-MRS) sequence with an optimized echo-time (97ms) on 3T-MRI. Spectroscopy data were analyzed using linear combination (LC) model. Cramer-Rao lower bound (CRLB) values of <40% were considered acceptable for detecting 2HG and <20% for other metabolites. Immunohistochemical analyses for determining IDH mutational status were subsequently performed from resected tumor specimens, and findings were compared with the results from spectral data. Mann-Whitney and Chi-squared tests were performed to ascertain differences in metabolite levels between IDH-mutant and IDH-wild-type gliomas. Receiver operating characteristics (ROC) curve analyses were also performed. Results Data from eight cases were excluded due to poor spectral quality or non-tumor-related etiology, and final data analyses were performed from 24 cases. Of these cases, 9/12 (75%) were correctly identified as IDH-mutant or IDH-wildtype gliomas by SVS and 10/12 (83%) by 1H-MRSI with an overall concordance rate of 79% (19/24). The sensitivity, specificity, positive predictive value, and negative predictive values were 80%, 77%, 86%, and 70%, respectively. The metabolite 2HG was found to be significant in predicting IDH-mutant gliomas by Chi-squared test (p<0.01). The IDH-mutant gliomas also showed significantly higher NAA/Cr ratio (1.20 ± 0.09 versus 0.75 ± 0.12 p= 0.016) and lower Glx/Cr ratio (0.86±0.078 versus 1.88±0.66; p=0.029) than those with IDH wild-type gliomas. The areas under the ROC curves for NAA/Cr and Glx/Cr were 0.808 and 0.786, respectively. Conclusion 1H-MRS may be useful in detecting IDH mutational status in gliomas by detecting 2HG and characteristic patterns of other metabolite ratios (NAA/Cr and Glx/Cr). This has significant clinical implications for prognostication and response assessment to novel targeted therapies in gliomas.