10,000 publications from this institution
Packet classification is a fundamental problem in computer networking. This problem exposes a hard tradeoff between the computation and state complexity, which makes it particularly challenging. To navigate this tradeoff, existing solutions rely on complex hand-tuned heuristics, which are brittle and hard to optimize. In this paper, we propose a deep reinforcement learning (RL) approach to solve the packet classification problem. There are several characteristics that make this problem a good fit for Deep RL. First, many of the existing solutions are iteratively building a decision tree by splitting nodes in the tree. Second, the effects of these actions (e.g., splitting nodes) can only be evaluated once we are done with building the tree. These two characteristics are naturally captured by the ability of RL to take actions that have sparse and delayed rewards. Third, it is computationally efficient to generate data traces and evaluate decision trees, which alleviate the notoriously high sample complexity problem of Deep RL algorithms. Our solution, NeuroCuts, uses succinct representations to encode state and action space, and efficiently explore candidate decision trees to optimize for a global objective. It produces compact decision trees optimized for a specific set of rules and a given performance metric, such as classification time, memory footprint, or a combination of the two. Evaluation on ClassBench shows that NeuroCuts outperforms existing hand-crafted algorithms in classification time by 18% at the median, and reduces both time and memory footprint by up to 3x.
SUMMARY. 1. Field experiments were conducted to examine the impact of grazing invertebrates on periphyton biomass in twenty‐one pools across three northern California coastal streams (U.S.A.): Big Sulphur Creek, the Rice Fork of the Eel River, and Big Canyon Creek. Periphyton accrual on artificial substrate tiles was compared in each stream between two treatments: those elevated slightly above the stream bottom to reduce access by grazers (= platforms) and those placed directly on the stream bottom to allow access by grazers (=controls). 2. Crawling invertebrate grazers (cased caddisflies and snails) were numerically dominant in each stream (86% of all grazers in Big Sulphur Creek, 61% in the Rice Fork, 84% in Big Canyon Creek). Platforms effectively excluded crawling grazers, but were less effective in excluding swimming mayfly grazers (Baetidae). 3. Periphyton biomass (as AFDM) on tiles was significantly lower on controls compared to platforms for the Rice Fork, an open‐canopy stream, and Big Sulphur Creek, a stream with a heterogeneous canopy. In contrast, no grazer impact was found for Big Canyon Creek, a densely shaded stream. Here, extremely low periphyton biomass occurred for both treatments throughout the 60 day study. 4. The influence of riparian canopy on periphyton growth (i.e. accrual on platforms), grazer impact on periphyton, and grazer abundance was examined for Big Sulphur Creek. As canopy increased (15–98% cover), periphyton biomass on platforms decreased. In contrast, canopy had little influence on periphyton accrual on controls; apparently, grazers could maintain low periphyton standing crops across the full range of canopy levels. The abundance of one grazer species, the caddisfly Gumaga nigricula , was highest in open, sunlit stream pools; abundance of two other prominent grazers, Helicopsyche borealis (Trichoptera) and Centroptilum convexum (Ephemeroptera), however, was unrelated to canopy.
Many large applications are now built using collections of microservices, each of which is deployed in isolated containers and which interact with each other through the use of remote procedure calls (RPCs). The use of microservices improves scalability -- each component of an application can be scaled independently -- and deployability. However, such applications are inherently distributed and current tools do not provide mechanisms to reason about and ensure their global behavior. In this paper we argue that recent advances in formal methods and software packet processing pave the path towards building mechanisms that can ensure correctness for such systems, both when they are being built and at runtime. These techniques impose minimal runtime overheads and are amenable to production deployments.
In this Perspective, we present progress, outstanding challenges, and opportunities for the incorporation of artificial metalloenzymes (ArMs) into biosynthetic pathways. We first explain discoveries within the field of ArMs that led to the potential inclusion of these enzymes in biosynthesis. We then describe the specific barriers that our laboratory, in collaboration with the laboratories of Keasling and Mukhopadhyay, addressed to establish a biosynthetic pathway containing an ArM. This biosynthesis produced an unnatural cyclopropyl terpenoid by combining heterologous production of the terpene with modification of its terminal alkene by an ArM built from a cytochrome P450. Finally, we describe the remaining challenges and opportunities related to the application of ArMs in synthetic biology.
We report the use of an amplified femtosecond laser for single-shot two-photon exposure of the commercial photoresist SU-8. By scanning of the focal volume through the interior of the resist, three-dimensional (3-D) structures are fabricated on a shot-by-shot basis. The 800-nm two-photon exposure and damage thresholds are 3.2 and 8.1TW/cm(2), respectively. The nonlinear nature of the two-photon process allows the production of features that are smaller than the diffraction limit. Preliminary results suggest that Ti:sapphire oscillators can achieve single-shot two-photon exposure with thresholds as low as 1.6TW/cm(2) at 700 nm, allowing 3-D structures to be constructed at megahertz repetition rates.
Linear spatial filtering is an important component of most image and video processing algorithms Therefore, when designing CNN Universal Machine (CNN-UM) algorithms for image and video applications, it would be useful to be able to implement desired filtering operations on the hardware. Although it has been shown that any convolution mask can, in principle, be implemented by a series of 3/spl times/3 template operations, such methods are time-consuming and error-prone. In this paper we investigate the use of simple CNN-UM algorithms involving only three filtering stages and using only 3/spl times/3 A- and B-templates to approximate desired filter transfer functions. The transfer functions for the structures are derived and a reduced parameterization is introduced. This form is conducive to optimization. Several examples are given wherein filters are designed to approximate a desired transfer function.
Much as Paul David described the invention of the mechanical typewriter – it was invented 51 times before being patented by Christopher Sholes in 1867, licensed to the Remington Company and successfully commercialized – the connections between the gold-exchange standard and the Great Depression have been discovered repeatedly. They were discovered by Ehsan Choudhri and Levis Kochin in a seminal article in 1980. They were discovered by Barry Eichengreen and Jeffrey Sachs in articles published in 1985 and 1986. They were discovered by James Hamilton in an insightful article published in 1988. They were discovered by Peter Temin in his Robbins Lectures published in 1989. They were discovered by the now chairman of the Federal Reserve Board in his 1994 Journal of Money Credit and Banking Lecture. Moreover, these contributors to the contemporary literature had important antecedents, including Robert Triffin in the 1950s, Ragnar Nurkse in the 1940s, and Leo Pasvolsky in the 1930s.
