לוח שנה משולב

Early warning systems are a class of risk prediction tools that have recently become part of the de facto approach towards improving high school graduation rates in the United States. These systems aim to help schools efficiently target resources to students by predicting which individuals are least likely to graduate, and hence need the most help.

In this talk, I will present the results of a collaboration with the Wisconsin Department of Public Instruction in which we conducted the first large-scale evaluation of the long-term impacts of early warning systems on high school graduation rates. Using a decade's worth of data and models, we find that risk assessments made by the system have been highly accurate at predicting student dropout, yet ineffective in improving outcomes. We will see how both of these findings can be simultaneously explained by the influence of structural, social factors. We will close with broader discussion regarding the broader policy implications of our work.

:The non-linearity and variability in individual mood responses pose multiple analytic and experimental challenges. These challenges limit our understanding of mental health disorders with aberrant mood dynamics such as depression, and the development of more effective treatments. Computational approaches can help overcome some of these challenges by creating and modeling individual mood transitions. I will describe a study where closed-loop control approach was used to generate individual mood transitions and then a computational modeling approach was used to characterize the temporal effects on these mood changes. This study showed that early events exert a stronger influence on reported mood compared to recent events (a primacy weighting), in contrary to previous theoretical accounts which assumed that recent events are most influential on mood. This Primacy model accounted better for mood reports compared to a range of alternative temporal representations, in random, consistent, or dynamic reward environments, across different age groups, and in both healthy and depressed participants. Moreover, I will show how this temporal relation between early experiences and mood is mediated by specific neural signals. Interestingly, in repetitive reward environments or resting-state conditions, we found that mood reports consistently decline over time, stressing the importance of accounting for temporal effects in mood responses. These findings hold implications for the timing of events when addressing mood and behavior in experimental and in clinical settings.

The discovery of planets capable of hosting biosignatures, and the characterization of the atmospheres of these planets, is a key and achievable goal in our lifetime. These goals require some of the most demanding precision spectroscopic and photometric measurements. I will discuss the instrumental challenges of detecting such planets with the Doppler radial velocity technique, and the evolution of the design of these instruments as they seek ever-tighter control of environmental parameters, and increased measurement precision. A suite of new technologies like frequency stabilized laser combs, low drift etalons, and deeper understanding of the detectors is enabling a new level of precision in radial velocity measurements - as well as illustrating new challenges. I will then discuss how the stars themselves are the remaining challenge, as magnetically driven processes create ‘stellar activity’ noise that can masquerade as planets and obfuscate their detection, and I highlight a few paths to mitigate this, along with some of the latest scientific results from the HPF and NEID instruments. I will discuss one iteration of a possible future, weaving its way from now through JWST individual and mini-population studies of planet atmospheres, large population studies with missions like ARIEL, the near-future of RV surveys, detection and characterization prospects with large ground-based, and the challenges and opportunities with future imaging and spectroscopic missions like LUVOIR and LIFE. The goal of discovering and characterizing terrestrial mass planets capable of hosting liquid water on their surfaces may now be within reach! But true understanding of the origin and meaning of the biosignatures we detect will likely require transdisciplinary research across multiple fields.

Transformer-based models have revolutionized natural language processing (NLP) and significantly improved various NLP tasks. However, many researchers make implicit assumptions about their training setups, assuming that the train and test sets are drawn from the same distribution. This assumption can limit the applicability of these models across different languages and domains.
The high cost of training state-of-the-art NLP models using various languages and domains has resulted in training them for only a subset of languages and domains, leading to a significant performance gap in excluded domains and languages. This performance gap marginalizes many individuals from accessing useful models.

This talk will address the challenges, approaches, and opportunities for democratizing NLP across different languages and domains.
Finally, we will explore future directions for making these models accessible to a broader audience.