Publications

This article presents the design, implementation, analysis, and experimental evaluation of speak-up , a defense against application-level distributed denial-of-service (DDoS), in which attackers cripple a server by sending legitimate-looking requests that consume computational resources (e.g., CPU cycles, disk). With speak-up, a victimized server encourages all clients, resources permitting, to automatically send higher volumes of traffic . We suppose that attackers are already using most of their upload bandwidth so cannot react to the encouragement. Good clients, however, have spare upload bandwidth so can react to the encouragement with drastically higher volumes of traffic. The intended outcome of this traffic inflation is that the good clients crowd out the bad ones, thereby capturing a much larger fraction of the server's resources than before. We experiment under various conditions and find that speak-up causes the server to spend resources on a group of clients in rough proportion to their aggregate upload bandwidths, which is the intended result.

Abstract Magnetic resonance imaging (MRI) contrast agents represent a worldwide billion‐dollar market annually. While T 1 relaxivity enhancement contrast agents receive greater attention and a significantly larger market share, the commercial potential for T 2 relaxivity enhancing contrast agents remains a viable diagnostic option because of their increased relaxivity at high field strengths. Improvement of the contrast and biocompatibility of T 2 MRI probes may enable new diagnostic prospects for MRI. Paramagnetic lanthanides have the potential to decrease T 1 and T 2 proton relaxation times, but are not commercially used in MRI diagnostics as T 2 agents. In this article, oxygen donor chelates (hydroxypyridinone, HOPO, and terephthalamide, TAM) of various lanthanides are demonstrated as biocompatible macromolecular dendrimer conjugates for the development of T 2 MRI probes. These conjugates have relaxivities of up to 374 m M –1 s –1 per dendrimer, high bioavailability, and low in vitro toxicity.

Syntheses of high-valent open-chain metallostannylene hydride complexes from transition metal polyhydride complexes and aryl-substituted amidostannylenes are reported. The reaction of (^MeBDI^Dipp)IrH₄ (^MeBDI^Dipp = (Dipp)NC(Me)CH(Me)CN(Dipp); Dipp = 2,6-diisopropylphenyl) with DMPSnN(SiMe₃)₂ (DMP = C₆H₃-2,6-Mes₂, Mes = mesityl) gives access to metallostannylene (^MeBDI^Dipp)IrH₃SnDMP (<b>1</b>) with HN(SiMe₃)₂ as the byproduct. Similarly, treatment of Cp*IrH₄ (Cp* = η⁵-C₅Me₅) with DMPSnN(SiMe₃)₂ leads to the formation of Cp*IrH₃SnDMP (<b>2</b>). In contrast to <b>1</b>, <b>2</b> cleanly reacts with a second equivalent of DMPSnN(SiMe₃)₂, allowing for the synthesis of bis(metallostannylene) Cp*IrH₂(SnDMP)₂ (<b>3</b>). Additionally, the synthesis and characterization of two new amidostannylenes DMPSn[N(Dipp)(SiMe₃)] (<b>4</b>) and TripSn[N(Dipp)(SiMe₃)] (<b>5</b>; Trip = 2,4,6-triisopropylphenyl) is presented. While the reaction of Cp*IrH₄ with <b>4</b> represents an alternative route to <b>2</b>, treatment of (^MeBDI^Dipp)IrH₄ with <b>5</b> yields the amidostannylene complex (^MeBDI^Dipp)(H)₂Ir═Sn(Trip){N(Dipp)(SiMe₃)} (<b>6</b>) with concomitant release of H₂, rendering stannylene coordination a reaction pathway that competes with Sn-N bond cleavage and metallostannylene formation. Within the broader context of transition metal-main group element compounds, metallostannylene formation is characterized as a formal deprotonation of the transition metal polyhydride complex by the amidostannylene. Notably, the ambiphilicity of amidostannylenes, which distinguishes them from tetravalent amidostannanes, appears to be essential for efficient Ir-Sn bond formation.

The high computational and memory requirements of large language model (LLM) inference make it feasible only with multiple high-end accelerators. Motivated by the emerging demand for latency-insensitive tasks with batched processing, this paper initiates the study of high-throughput LLM inference using limited resources, such as a single commodity GPU. We present FlexGen, a high-throughput generation engine for running LLMs with limited GPU memory. FlexGen can be flexibly configured under various hardware resource constraints by aggregating memory and computation from the GPU, CPU, and disk. By solving a linear programming problem, it searches for efficient patterns to store and access tensors. FlexGen further compresses the weights and the attention cache to 4 bits with negligible accuracy loss. These techniques enable FlexGen to have a larger space of batch size choices and thus significantly increase maximum throughput. As a result, when running OPT-175B on a single 16GB GPU, FlexGen achieves significantly higher throughput compared to state-of-the-art offloading systems, reaching a generation throughput of 1 token/s for the first time with an effective batch size of 144. On the HELM benchmark, FlexGen can benchmark a 30B model with a 16GB GPU on 7 representative sub-scenarios in 21 hours. The code is available at https://github.com/FMInference/FlexGen

The effect of alloying small amounts of ZnO with ZnSe on the electronic band structure has been studied. Optical transitions in molecular-beam-epitaxy-grown ZnOxSe1−x epitaxial films (0⩽x⩽1.35%) were investigated using photoreflectance and photoluminescence spectroscopies. The fundamental band-gap energy of the alloys was found to decrease at a rate of about 0.1 eV per atomic percent of oxygen. The pressure dependence of the band gap was also found to be strongly affected by O incorporation. Both the effects can be quantitatively explained by an anticrossing interaction between the extended states of the conduction band of ZnSe and the highly localized oxygen states located at approximately 0.22 eV above the conduction-band edge.