Papers by Huynh My Nhi (BTEC HCM)
Biophysical Journal, 2008
We investigated the effect of combined inhibition of oxidative and glycolytic metabolism on L-typ... more We investigated the effect of combined inhibition of oxidative and glycolytic metabolism on L-type Ca 21 channels (LCCs) and Ca 21 spikes in isolated patch-clamped rabbit ventricular myocytes. Metabolic inhibition (MI) reduced LCC open probability, increased null probability, increased first latency, and decreased open time but left conductance unchanged. These results explain the reduction in macroscopic Ca 21 current observed during MI. MI also produced a gradual reduction in action potential duration at 90% repolarization (APD 90 ), a clear decline in spike probability, and an increase in spike latency and variance. These effects are consistent with the changes we observed in LCC activity. MI had no effect on the amplitude or time to peak of Ca 21 spikes until APD 90 reached 10% of control, suggesting preserved sarcoplasmic reticulum Ca 21 stores and ryanodine receptor (RyR) conductance in those couplons that remained functioning. Ca 21 spikes disappeared completely when APD 90 reached ,2% of control, although in two cells, spikes were reactivated in a highly synchronized fashion by very short action potentials. This reactivation is probably due to the increased driving force for Ca 21 entry through a reduced number of LCCs that remain open during early repolarization. The enlarged single channel flux produced by rapid repolarization is apparently sufficient to trigger RyRs whose Ca 21 sensitivity is likely reduced by MI. We suggest that loss of coupling fidelity during MI is explained by loss of LCC activity (possibly mediated by Ca 21 -calmodulin kinase II activity). In addition, the results are consistent with loss of RyR activity, which can be mitigated under conditions likely to enlarge the trigger. We isolated ventricular myocytes from adult New Zealand White rabbits (2.0 to 3.0 kg) using collagenase/protease digestion as described previously
arXiv (Cornell University), Jul 14, 2021
Deep generative models (DGMs) seem a natural fit for detecting out-of-distribution (OOD) inputs, ... more Deep generative models (DGMs) seem a natural fit for detecting out-of-distribution (OOD) inputs, but such models have been shown to assign higher probabilities or densities to OOD images than images from the training distribution. In this work, we explain why this behavior should be attributed to model misestimation. We first prove that no method can guarantee performance beyond random chance without assumptions on which out-distributions are relevant. We then interrogate the typical set hypothesis, the claim that relevant out-distributions can lie in high likelihood regions of the data distribution, and that OOD detection should be defined based on the data distribution's typical set. We highlight the consequences implied by assuming support overlap between in-and out-distributions, as well as the arbitrariness of the typical set for OOD detection. Our results suggest that estimation error is a more plausible explanation than the misalignment between likelihood-based OOD detection and outdistributions of interest, and we illustrate how even minimal estimation error can lead to OOD detection failures, yielding implications for future work in deep generative modeling and OOD detection.
arXiv (Cornell University), May 16, 2021
Typically, information extraction (IE) requires a pipeline approach: first, a sequence labeling m... more Typically, information extraction (IE) requires a pipeline approach: first, a sequence labeling model is trained on manually annotated documents to extract relevant spans; then, when a new document arrives, a model predicts spans which are then post-processed and standardized to convert the information into a database entry. We replace this labor-intensive workflow with a transformer language model trained on existing database records to directly generate structured JSON. Our solution removes the workload associated with producing token-level annotations and takes advantage of a data source which is generally quite plentiful (e.g. database records). As long documents are common in information extraction tasks, we use gradient checkpointing and chunked encoding to apply our method to sequences of up to 32,000 tokens on a single GPU. Our Doc2Dict approach is competitive with more complex, hand-engineered pipelines and offers a simple but effective baseline for documentlevel information extraction. We release our Doc2Dict model and code to reproduce our experiments and facilitate future work. 1
In many prediction problems, spurious correlations are induced by a changing relationship between... more In many prediction problems, spurious correlations are induced by a changing relationship between the label and a nuisance variable that is also correlated with the covariates. For example, in classifying animals in natural images, the background, which is the nuisance, can predict the type of animal. This nuisance-label relationship does not always hold, and the performance of a model trained under one such relationship may be poor on data with a different nuisance-label relationship. To build predictive models that perform well regardless of the nuisance-label relationship, we develop Nuisance-Randomized Distillation (NURD). We first define the nuisance-varying family, a set of distributions that differ only in the nuisance-label relationship. We then introduce the nuisance-randomized distribution, a distribution where the nuisance and the label are independent. Under this distribution, we define the set of representations such that conditioning on any member, the nuisance and the...
American journal of public health, 2015
In an antifluoridation case study, we explored digital pandemics and the social spread of scienti... more In an antifluoridation case study, we explored digital pandemics and the social spread of scientifically inaccurate health information across the Web, and we considered the potential health effects. Using the social networking site Facebook and the open source applications Netvizz and Gephi, we analyzed the connectedness of antifluoride networks as a measure of social influence, the social diffusion of information based on conversations about a sample scientific publication as a measure of spread, and the engagement and sentiment about the publication as a measure of attitudes and behaviors. Our study sample was significantly more connected than was the social networking site overall (P<.001). Social diffusion was evident; users were forced to navigate multiple pages or never reached the sample publication being discussed 60% and 12% of the time, respectively. Users had a 1 in 2 chance of encountering negative and nonempirical content about fluoride unrelated to the sample public...
Uploads
Papers by Huynh My Nhi (BTEC HCM)