Diffusion Transformers (DiTs) incur high memory traffic and energy costs because sampling repeatedly evaluates large denoisers dominated by linear operations. Analog compute-in-memory (CIM) can alleviate these costs by executing linear operations within weight-storing memory arrays. However, CIM nonidealities perturb effective weights, with errors accumulating along the state-dependent denoising trajectory; their interaction with classifier-free guidance (CFG) remains underexplored. In this paper, we characterize the impact of analog CIM nonidealities on DiT sampling. Although conditional and unconditional predictions can each remain close to their clean counterparts, their difference (the CFG residual) is disproportionately attenuated and rotated. Identifying this residual as a controllable failure channel, we propose a retraining-free, sampler-side recalibration that adjusts only the CFG scale for a given CIM condition. Trajectory-level analysis shows that moderate recalibration strengthens the target-oriented component preserved in the distorted residual, enabling earlier commitment to a prompt-consistent semantic region. In contrast, excessive guidance amplifies the full noisy residual and degrades quality, resulting in a finite, noise-dependent optimum. Extensive experiments on PixArt-Sigma, PixArt-alpha, and DiT-XL/2 show that the optimal guidance scale increases with CIM noise. Using 30,000 samples per condition, guidance recalibration consistently restores generation quality across simulated CIM mappings, closing at least 87% of the CIM-induced FID gap at a CIM noise level of 0.20. It reduces FID from 59.22 to 20.49 on PixArt-Sigma, 72.37 to 21.12 on PixArt-alpha, and 20.89 to 6.62 on DiT-XL/2.
Abstract Background Acute respiratory infections (ARIs) remain a major global health concern. Although long-term air pollution exposure has been linked to ARIs, prospective evidence from community-based populations remains limited. Objective This study aimed to quantify the burden of ARIs in the community and examine the associations between long-term exposure to particulate matter 2.5 (PM2.5) and ozone (O3) and the risks of ARIs, with additional analyses using influenza-like illness (ILI) as a more specific outcome. Methods We conducted a prospective cohort study including 3617 residents in Shanghai, China, who were followed weekly for 1 year. Individual-level exposure to PM2.5 and O3 concentrations was estimated using high-resolution datasets. Cox proportional hazards models with shared frailty were applied to assess associations with ARIs. Exposure windows of 3, 6, 9, and 12 months were evaluated, and the optimal window was selected based on the Akaike information criterion. Effect estimates were reported per IQR increase. Dose-response relationships, subgroup analyses, and multiple sensitivity analyses were performed. Results During 3217 person-years of follow-up, 885 ARI events were documented (0.27 per person-year). In the fully adjusted model using the 12-month exposure window, each IQR increase in PM2.5 was associated with higher risks of ARIs (hazard ratio [HR] 1.594, 95% CI 1.340‐1.897), with stronger associations observed for ILI (HR 1.948, 95% CI 1.484‐2.557). For O3, the corresponding HRs were 1.510 (95% CI 1.135‐2.007) for ARIs, with stronger associations for ILI (HR 2.229, 95% CI 1.385‐3.588). PM2.5 showed a nonlinear association with ARIs, whereas linear relationships were observed for PM2.5 with ILI and O3 with both outcomes. Evidence of effect modification was observed by age, residence, and season for PM2.5 and by season for O3. Results were robust across multiple sensitivity analyses. Conclusions Long-term exposure to PM2.5 and O3 is associated with increased risk of ARIs, with similar associations observed for ILI. These findings highlight the importance of long-term air pollution control and targeted interventions for susceptible populations, particularly during cold seasons.
Xiaole Duan, Xiao Yu, Hongyu Liang et al.· JMIR Public Health and Surve...· 0 citations
Approximate Speculative Decoding (ASD) is introduced, a training-free verifier that replaces binary first-mismatch truncation with budgeted longest-prefix selection and reuses the contiguous target-greedy suffix without additional approximate decisions or target-model forward passes.
Yuannuo Feng, Zegang Peng, Yuxin Xie et al.· 0 citations