Integrated Calendar

N2 fixation is the primary pathway by which bioavailable nitrogen is added to theoceans. However, the drivers of N2 fixation on orbital timescales are uncertain. Wepresent high-resolution foraminifera-bound (FB) δ15N records from the Westernand Eastern tropical Atlantic Ocean (WTA and ETA respectively) throughout thelate Pliocene (~3.60 to ~1.97 Ma), where WTA ODP Site 999 represents N2fixation changes and EEA ODP Site 662 represents changes in pycnocline δ15N.Our results show that, compared to the past 160 ka, N2 fixation in the WTA wassignificantly lower throughout the late Pliocene as reflected by an average of ~2 ‰higher FB-δ15N values. A possible explanation to the higher Pliocene FB-δ15N inthe WTA could be lower rates of global denitrification that were balanced by lowerglobal N2 fixation levels. We suggest that this reduced N2 fixation was due todecreased excess P in the pycnocline/subsurface ocean, driven by lower globalwater column denitrification. This finding implies a coupling between decreasedwater column denitrification and reduced level N2 fixation rates under warmerclimates.On orbital timescales, our N2 fixation record display obliquity-paced cycles thatprogressively intensified after the Northern Hemisphere glaciation intensification ~2.8 Ma, and the onset of equatorial upwelling pulses documented during glacialperiods in the EEA (ODP Site 662; [1]). The observed changes in N2 fixation of thelast 160 ka were previously explained by precession-paced upwelling in the EEAthat imported excess P into the oligotrophic WTA [2]. However, precessionalcyclicity is not dominant in the Pliocene FB- δ15N, which calls for other candidatesto explain the variations after 2.8 Ma. The best explanation is a response to sealevelpaced sedimentary denitrification. Glacial lower sea levels exposedcontinental shelves, reducing regional benthic denitrification and inhibiting thesupply of excess P, thereby limiting N2 fixation in the WTA, whereas interglacialsubmerged shelves increased excess P availability.

Given two distributions P, Q and an integer t, we analyze two sampling processes. In "random allocation," we first sample an index i uniformly from [t], then draw r_{i} ~ P and r_{j} ~ Q for all other j in [t]. In "Poisson sampling," we independently draw r_{i} ~ 1/t*P + (1-1/t)*Q for each i in [t]. We bound the difference between these processes' output distributions and the baseline of sampling r_{i} ~ Q for all i.

This theoretical result provides key insights for analyzing DP-SGD, a privacy-preserving variant of stochastic gradient descent. While Poisson subsampling has well-understood privacy guarantees, common implementations use element shuffling, which was recently shown to have larger privacy losses in certain regimes. Random allocation offers a middle ground, and we prove its privacy analysis reduces to comparing the distributions described above.

We show that these variants' privacy guarantees are within a constant factor of each other across all parameter regimes and converge asymptotically in t. Our proof has two key components: decomposing Poisson sampling into a mixture of random allocation processes, and showing that random allocation can be viewed as a modified Poisson process where sampling probabilities depend on previous outputs.

Joint work with Vitaly Feldman

Moiré materials built on transition metal dichalcogenide semiconductors have emerged as a tunable platform for simulating the Hubbard model on a triangular lattice. A natural question arises: Can the platform be tuned to yield a phase diagram similar to that in high-temperature cuprate superconductors? In this talk, I will discuss the emergence of “high-temperature” superconductivity near the Mott transition in a triangular moiré lattice with intermediate coupling strength. The emergent doping-temperature phase diagram looks remarkably similar to that in cuprate superconductors. I will also discuss the evolution of the phase diagram by tuning the band structure of the material by gating. The results could provide a new angle for understanding the phenomenon of high-temperature superconductivity in strongly correlated materials.  

Understanding why some patients respond to immune checkpoint therapy while others do not remains a critical challenge in cancer research. This talk will explore three key studies that shed light on this question. First, we uncover metabolic predictors of response to checkpoint blockade therapy, revealing how tumor and immune cell metabolism shape treatment outcomes. Next, we present a single-cell meta-analysis of T cell clonal dynamics, highlighting their role in immunotherapy success. Finally, we introduce scXpand, a machine-learning approach for predicting T cell clonality from scRNA-seq, offering a novel tool to enhance immunotherapy research and precision medicine.

Given a distribution of images, how can we can decompose the data into a set of underlying components? In this talk, I'll present an approach that decomposes images into a underlying composable energy functions. I'll illustrate how energy functions allow us to represent both global components of an image, such as lighting as well as local components such as objects. I'll further show how we leverage pretrained vision models to infer these components. Finally, I'll illustrate how discover components can be recombined to form a variety of images substantially different than those seen at training time.

Speaker's bio:

Yilun Du is an incoming assistant professor at Harvard and is currently a senior research scientist at Google Deepmind. He received has PhD and BS from MIT and was supported by a NSF graduate fellowship.