10,000 publications from this institution
Reinforcement Learning (RL) algorithms are often known for sample inefficiency and difficult generalization. Recently, Unsupervised Environment Design (UED) emerged as a new paradigm for zero-shot generalization by simultaneously learning a task distribution and agent policies on the generated tasks. This is a non-stationary process where the task distribution evolves along with agent policies; creating an instability over time. While past works demonstrated the potential of such approaches, sampling effectively from the task space remains an open challenge, bottlenecking these approaches. To this end, we introduce CLUTR: a novel unsupervised curriculum learning algorithm that decouples task representation and curriculum learning into a two-stage optimization. It first trains a recurrent variational autoencoder on randomly generated tasks to learn a latent task manifold. Next, a teacher agent creates a curriculum by maximizing a minimax REGRET-based objective on a set of latent tasks sampled from this manifold. Using the fixed-pretrained task manifold, we show that CLUTR successfully overcomes the non-stationarity problem and improves stability. Our experimental results show CLUTR outperforms PAIRED, a principled and popular UED method, in the challenging CarRacing and navigation environments: achieving 10.6X and 45\% improvement in zero-shot generalization, respectively. CLUTR also performs comparably to the non-UED state-of-the-art for CarRacing, while requiring 500X fewer environment interactions.
The Siberian Traps represent one of the most voluminous flood basalt provinces on Earth. Laser-heating 40 Ar/ 39 Ar data indicate that the bulk of these basalts was erupted over an extremely short time interval (900,000 ± 800,000 years) beginning at about 248 million years ago at mean eruption rates of greater than 1.3 cubic kilometers per year. Such rates are consistent with a mantle plume origin. Magmatism was not associated with significant lithospheric rifting; thus, mantle decompression resulting from rifting was probably not the primary cause of widespread melting. Inception of Siberian Traps volcanism coincided (within uncertainty) with a profound faunal mass extinction at the Permo-Triassic boundary 249 ± 4 million years ago; these data thus leave open the question of a genetic relation between the two events.
An entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available from the CCDC and typically includes 3D coordinates, cell parameters, space group, experimental conditions and quality measures.
Abstract For Abstract see ChemInform Abstract in Full Text.
— The use of d.c. electrical potential methods is described for the monitoring of Mode III (anti‐plane shear) fatigue cracks in circumferentially‐notched cylindrical specimens subjected to cyclic torsion. Calibration of potential change with crack depth and optimization of current input and potential measurement probe locations are achieved using simple finite element procedures, and are verified experimentally. The use of the method for Mode III crack growth studies is described in the light of crack face electrical shorting problems associated with torsional crack closure.
Analog arrays are a generalization of cellular neural networks (CNN) which consist of a regular array of nonlinear analog processors at each node and nearest neighbor interactions. Analog arrays can incorporate nonlinearities in both the input and output functions and, contrasted with CNN arrays, have continuous-valued outputs in the equilibrium state. The general analog array, though more powerful than the CNN, presents a functional test nightmare. Since the output is continuous-valued (even at equilibrium) and the dynamics can be complicated, evaluating whether a fabricated VLSI array complies with the intended processing function for a wide range of inputs can be very difficult and time consuming. The number of analog inputs and outputs also strains most modern analog VLSI automatic test equipment (ATE). The authors present both a new hardware design for massive analog circuit testing and algorithms for functional test of analog arrays.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">></ETX>
The need for real-time processing of "big data" has led to the development of frameworks for distributed stream processing in clusters. It is important for such frameworks to be robust against variable operating conditions such as server failures, changes in data ingestion rates, and workload characteristics. To provide fault tolerance and efficient stream processing at scale, recent stream processing frameworks have proposed to treat streaming workloads as a series of batch jobs on small batches of streaming data. However, the robustness of such frameworks against variable operating conditions has not been explored.
The gas-phase carbonylation of dimethoxymethane (DMM) to form methyl methoxyacetate (MMAc) can be catalyzed by acid zeolites. This reaction is a critical step in the synthesis of monoethylene glycol (MEG), a widely used chemical, from synthesis gas. The mechanism of DMM carbonylation occurring on H−MFI and H−FAU zeolites has been investigated using density functional theory. We find that the reaction involves three steps: initiation via reaction of zeolite protons with DMM to form methoxymethoxy species, carbonylation of the resulting species, and subsequent methoxylation of the resulting acyl species. Both the carbonylation and methoxylation processes proceed via carbocationic transition states that are stabilized by the framework O atoms of the zeolite. The activation barriers for carbonylation are similar in both zeolites, but the barriers for methoxylation differ significantly. Energy decomposition analysis indicates that a combination of the pore size and of the flexibility of the reactive species gives rise to the differences in reactivity between the zeolites. The effect of basis set superposition was assessed using a 6-311++G(3df,3pd) basis set. This effect depends strongly on the gas-phase molecules involved but very weakly on the zeolite framework, and its estimate can be transferred from one zeolite to another to reduce the computational expense of such simulations.
