arXiv subjects

Zihan Zhang

Publications and source records attributed to Zihan Zhang.

211 records · Page 12Linked to original sources

arXiv2020

Model-Free Reinforcement Learning: from Clipped Pseudo-Regret to Sample Complexity

In this paper we consider the problem of learning an $ε$-optimal policy for a discounted Markov Decision Process (MDP). Given an MDP with $S$ states, $A$ actions, the discount factor $γ\in (0,1)$, and an approximation threshold $ε> 0$, we provide a model-free algorithm to learn an $ε$-optimal policy with sample complexity $\tilde{O}(\frac{SA\ln(1/p)}{ε^2(1-γ)^{5.5}})$ (where the notation $\tilde{O}(\cdot)$ hides poly-logarithmic factors of $S,A,1/(1-γ)$, and $1/ε$) and success probability $(1-p)$. For small enough $ε$, we show an improved algorithm with sample complexity $\tilde{O}(\frac{SA\ln(1/p)}{ε^2(1-γ)^{3}})$. While the first bound improves upon all known model-free algorithms and model-based ones with tight dependence on $S$, our second algorithm beats all known sample complexity bounds and matches the information theoretic lower bound up to logarithmic factors.

cs.LG↗

arXiv2020

New Phase Transition of Lanthanum at High Pressure

Lanthanum (La), the first member of the rare-earth elements, recently aroused strong interest due to its unique superhydride with superconducting properties. Although there is much theoretical and experimental work about phase transitions and superconductivity in metallic La, we got a new body-centred tetragonal ($bct$) phase in metallic La with space group $I4/mmm$ at $190$ GPa exchanging from face-centered cubic ($fcc$) phase in previous work, which expanded the phase transition sequence. The $bct$ phase shows an abnormal packing way that turned to non-closed packing at high pressure. And more detailed properties of the new phase are discussed.

cond-mat.mtrl-sci↗

arXiv2020

Superionic State Discovered in Ternary Hypervalent Silicon Hydrides via Sodium inside the Earth

Superionic states are phases of matters that can simultaneously exhibit some of the properties of a fluid and of a solid. Superionic states of ice, H$_{3}$O, He-H$_{2}$O or He-NH$_{3}$ compounds have been reported in previous works. Silicon, sodium, and hydrogen are abundant elements inside the earth. Here, we use ab initio calculations to show that, at extreme conditions inside the earth, Na, Si, and H can form many hypervalent compounds that some of them can exist every close to ambient pressure, and surprisingly a previously unknown type of superionic state of $P\overline3m1 - $Na$_{2}$SiH$_{6}$ can form as well. Our work focused on new superionic state of Na$_{2}$SiH$_{6}$, and the results also reveal several different hypervalent Si-H anions discovered, which are different from individual SiH$_{5}^{\,-}$ and octahedral SiH$_{6}^{\,2-}$ in previous research of ternary alkali hypervalent silicon hydrides. Our work provides some advice on further investigations on potential ternary hydrides inside the earth.

cond-mat.mtrl-sci↗

arXiv2020

Nearly Minimax Optimal Reward-free Reinforcement Learning

We study the reward-free reinforcement learning framework, which is particularly suitable for batch reinforcement learning and scenarios where one needs policies for multiple reward functions. This framework has two phases. In the exploration phase, the agent collects trajectories by interacting with the environment without using any reward signal. In the planning phase, the agent needs to return a near-optimal policy for arbitrary reward functions. We give a new efficient algorithm, \textbf{S}taged \textbf{S}ampling + \textbf{T}runcated \textbf{P}lanning (\algoname), which interacts with the environment at most $O\left( \frac{S^2A}{ε^2}\text{poly}\log\left(\frac{SAH}ε\right) \right)$ episodes in the exploration phase, and guarantees to output a near-optimal policy for arbitrary reward functions in the planning phase. Here, $S$ is the size of state space, $A$ is the size of action space, $H$ is the planning horizon, and $ε$ is the target accuracy relative to the total reward. Notably, our sample complexity scales only \emph{logarithmically} with $H$, in contrast to all existing results which scale \emph{polynomially} with $H$. Furthermore, this bound matches the minimax lower bound $Ω\left(\frac{S^2A}{ε^2}\right)$ up to logarithmic factors. Our results rely on three new techniques : 1) A new sufficient condition for the dataset to plan for an $ε$-suboptimal policy; 2) A new way to plan efficiently under the proposed condition using soft-truncated planning; 3) Constructing extended MDP to maximize the truncated accumulative rewards efficiently.

cs.LG↗

arXiv2020

High Tc superconductivity in heavy Rare Earth Hydrides: correlation between the presence of the f states on the Fermi surface, nesting and the value of Tc

Lanthanum hydrides, containing hydrogen framework structures under compression, display a superconducting state with a high observed critical temperature. However, this phenomenon has so far only been observed at very high pressures. Here, we computationally search for superconductors with very high critical temperatures, but at much lower pressures. We uncover two such sodalite-type clathrate hydrides, YbH6 and LuH6 (Tc =145 K at P=70 GPa for YbH6 and, especially, Tc =273 K at P=100 GPa for LuH6). These striking properties are a consequence of the strong interrelationship between the f states present at the Fermi surface, structural stability and Tc value. For example, TmH6, with unfilled 4f orbitals, is stable at 50 GPa, while it has a relatively low value of Tc of 25 K. As for the YbH6 and LuH6 compounds, they have filled f-shells and the decrease of the f-energy below the Fermi level leads to formation of the nesting regions on the Fermi surface and, as a result, to phonon "softening" and an increase in Tc.

cond-mat.supr-con↗

arXiv2020

Ternary Hypervalent Silicon Hydrides via Lithium at High Pressure

Hydrogen is rarely observed as ligand in hypervalent species, however, we find that high-pressure hydrogenation may stabilise hypervalent hydrogen-rich materials. Focussing on ternary silicon hydrides via lithium doping, we find anions composed of hypervalent silicon with H ligands formed under high pressure. Our results reveal two new hypervalent anions: layered-SiH$_{5}^{-}$ and tricapped trigonal prismatic SiH$_{6}^{2-}$. These differ from octahedral SiH$_{6}^{2-}$ described in earlier studies. In addition, there are further hydrogen-rich structures Li$_{3}$SiH$_{10}$ and Li$_{2}$SiH$_{6+δ}$ which may be stabilised at high pressure. Our work provides pointers to future investigations on hydrogen-rich materials.

cond-mat.mtrl-sci↗

arXiv2020

Almost Optimal Model-Free Reinforcement Learning via Reference-Advantage Decomposition

We study the reinforcement learning problem in the setting of finite-horizon episodic Markov Decision Processes (MDPs) with $S$ states, $A$ actions, and episode length $H$. We propose a model-free algorithm UCB-Advantage and prove that it achieves $\tilde{O}(\sqrt{H^2SAT})$ regret where $T = KH$ and $K$ is the number of episodes to play. Our regret bound improves upon the results of [Jin et al., 2018] and matches the best known model-based algorithms as well as the information theoretic lower bound up to logarithmic factors. We also show that UCB-Advantage achieves low local switching cost and applies to concurrent reinforcement learning, improving upon the recent results of [Bai et al., 2019].

cs.LG↗

arXiv2020

Hydrogen "penta-graphene-like" structure stabilized by hafnium: a high-temperature conventional superconductor

The recent discovery of H3S and LaH10 superconductors with record high superconducting transition temperatures, Tc, at high pressure, has fueled the search for room-temperature superconductivity in the compressed superhydrides. Here we predict the existence of an unprecedented hexagonal HfH10, with an extraordinarily high Tc of around 213-234 K at 250 GPa. In HfH10, the H atoms are arranged in clusters to form a planar "penta-graphene-like" sublattice, in contrast to the covalent sixfold cubic structure in H3S and clathrate-like structure in LaH10. The Hf atom acts as a "precompressor" and electron donor to the hydrogen sublattice. This "penta-graphene-like" H10 structure is also found in ZrH10, ScH10 and LuH10 at high pressure, each material showing a high Tc ranging from 134 to 220 kelvin. Our study of dense superhydrides with "penta-graphene-like" layered structures opens the door to the exploration and exploitation of a new class of high Tc superconductors.

cond-mat.supr-con↗

arXiv2019

Regret Minimization for Reinforcement Learning by Evaluating the Optimal Bias Function

We present an algorithm based on the \emph{Optimism in the Face of Uncertainty} (OFU) principle which is able to learn Reinforcement Learning (RL) modeled by Markov decision process (MDP) with finite state-action space efficiently. By evaluating the state-pair difference of the optimal bias function $h^{*}$, the proposed algorithm achieves a regret bound of $\tilde{O}(\sqrt{SAHT})$\footnote{The symbol $\tilde{O}$ means $O$ with log factors ignored. } for MDP with $S$ states and $A$ actions, in the case that an upper bound $H$ on the span of $h^{*}$, i.e., $sp(h^{*})$ is known. This result outperforms the best previous regret bounds $\tilde{O}(S\sqrt{AHT}) $\citep{fruit2019improved} by a factor of $\sqrt{S}$. Furthermore, this regret bound matches the lower bound of $Ω(\sqrt{SAHT}) $\citep{jaksch2010near} up to a logarithmic factor. As a consequence, we show that there is a near optimal regret bound of $\tilde{O}(\sqrt{SADT})$ for MDPs with a finite diameter $D$ compared to the lower bound of $Ω(\sqrt{SADT}) $\citep{jaksch2010near}.

cs.LG↗

arXiv2019

AttentionXML: Label Tree-based Attention-Aware Deep Model for High-Performance Extreme Multi-Label Text Classification

Extreme multi-label text classification (XMTC) is an important problem in the era of big data, for tagging a given text with the most relevant multiple labels from an extremely large-scale label set. XMTC can be found in many applications, such as item categorization, web page tagging, and news annotation. Traditionally most methods used bag-of-words (BOW) as inputs, ignoring word context as well as deep semantic information. Recent attempts to overcome the problems of BOW by deep learning still suffer from 1) failing to capture the important subtext for each label and 2) lack of scalability against the huge number of labels. We propose a new label tree-based deep learning model for XMTC, called AttentionXML, with two unique features: 1) a multi-label attention mechanism with raw text as input, which allows to capture the most relevant part of text to each label; and 2) a shallow and wide probabilistic label tree (PLT), which allows to handle millions of labels, especially for "tail labels". We empirically compared the performance of AttentionXML with those of eight state-of-the-art methods over six benchmark datasets, including Amazon-3M with around 3 million labels. AttentionXML outperformed all competing methods under all experimental settings. Experimental results also show that AttentionXML achieved the best performance against tail labels among label tree-based methods. The code and datasets are available at http://github.com/yourh/AttentionXML .

cs.CL↗

arXiv2019

HAXMLNet: Hierarchical Attention Network for Extreme Multi-Label Text Classification

Extreme multi-label text classification (XMTC) addresses the problem of tagging each text with the most relevant labels from an extreme-scale label set. Traditional methods use bag-of-words (BOW) representations without context information as their features. The state-ot-the-art deep learning-based method, AttentionXML, which uses a recurrent neural network (RNN) and the multi-label attention, can hardly deal with extreme-scale (hundreds of thousands labels) problem. To address this, we propose our HAXMLNet, which uses an efficient and effective hierarchical structure with the multi-label attention. Experimental results show that HAXMLNet reaches a competitive performance with other state-of-the-art methods.

cs.IR↗

arXiv2019

Hybrid-Functional Calculations of Electronic Structure and Phase Stability of MO (M = Zn, Cd, Be, Mg, Ca, Sr, Ba) and Related Ternary Alloy M$_x$Zn$_{1-x}$O

Using the hybrid exchange-correlation functional within the density-functional theory, we have systematically investigated the structural and electronic properties of MO (M = Be, Mg, Ca, Sr, Ba, Zn, Cd) in binary rocksalt (B1), zincblende (B3) and wurtzite (B4) phases, including the structural parameters, bulk moduli, band gaps and deformation potentials. Our results agree well with the experimental data and other theoretical results, and give a better understanding of the relationship between the geometric and electronic structure. After calculating the band alignment, we find that in both the B1 and B3 structures, the valence band maximum (VBM) has an obvious decrease from BeO to MgO to CaO, then it goes up from SrO to BaO to ZnO to CdO. Moreover, the properties of the ternary alloys M$_x$Zn$_{1-x}$O were studied through the application of the special quasirandom structure method. The critical value of the ZnO composition for the transition from the B3 structure to the B1 structure gradually increases from (Ca, Zn)O to (Mg, Zn)O to (Sr, Zn)O to (Ba, Zn)O to (Cd, Zn)O, indicating that (Ca, Zn)O can exist in the B3 structure with the lowest ZnO composition. These results provide a good guideline for the accessible phase space in these alloy systems.

cond-mat.mtrl-sci↗

arXiv2018

Competing dynamics of single phosphorus dopant in graphene with electron irradiation

Atomic-level structural changes in materials are important but challenging to study. Here, we demonstrate the dynamics and the possibility of manipulating a phosphorus dopant atom in graphene using scanning transmission electron microscopy (STEM). The mechanisms of various processes are explored and compared with those of other dopant species by first-principles calculations. This work paves the way for designing a more precise and optimized protocol for atomic engineering.

cond-mat.mtrl-sci↗