BZPEERSource author record
Explore this author’s publications, collaborators and research topics. This catalog record has not been claimed by its author.
Catalog footprint
Published work
We consider the convergence theory for dyadic approximation in the middle-third Cantor set, K, for approximation functions of the form ψτ(n)=n−τ (τ≥0). In particular, we show that for values of τ beyond a certain threshold we have that almost no point in K is dyadically ψτ-well approximable with respect to the natural probability measure on K. This refines a previous result in this direction obtained by the first, third, and fourth named authors (arXiv, 2020).
Let C be the middle third Cantor set and μ be the log3log2-dimensional Hausdorff measure restricted to C. In this paper we study approximations of elements of C by dyadic rationals. Our main result implies that for μ almost every x∈C we have #{1≤n≤N:x−2np≤n0.01⋅2n1 for some p∈N}∼2∑n=1Nn−0.01. This improves upon a recent result of Allen, Chow, and Yu which gives a sub-logarithmic improvement over the trivial approximation rate.
A well known theorem due to Koksma states that for Lebesgue almost every x>1 the sequence (xn)n=1∞ is uniformly distributed modulo one. In this paper we give sufficient conditions for an analogue of this theorem to hold for self-similar measures. Our approach applies more generally to sequences of the form (fn(x))n=1∞ where (fn)n=1∞ is a sequence of sufficiently smooth real valued functions satisfying a nonlinearity assumption. As a corollary of our main result, we show that if C is equal to the middle third Cantor set and t≥1, then with respect to the Cantor-Lebesgue measure on C+t the sequence (xn)n=1∞ is uniformly distributed for almost every x.
In this paper we obtain new lower bounds for the upper box dimension of αβ sets. As a corollary of our main result, we show that if α is not a Liouville number and β is a Liouville number, then the upper box dimension of any αβ set is 1. We also use our dimension bounds to obtain new results on affine embeddings of self-similar sets.
Non-autoregressive generation (NAG) has recently attracted great attention due to its fast inference speed. However, the generation quality of existing NAG models still lags behind their autoregressive counterparts. In this work, we show that BERT can be employed as the backbone of a NAG model to greatly improve performance. Additionally, we devise mechanisms to alleviate the two common problems of vanilla NAG models: the inflexibility of prefixed output length and the conditional independence of individual token predictions. Lastly, to further increase the speed advantage of the proposed model, we propose a new decoding strategy, ratio-first, for applications where the output lengths can be approximately estimated beforehand. For a comprehensive evaluation, we test the proposed model on three text generation tasks, including text summarization, sentence compression and machine translation. Experimental results show that our model significantly outperforms existing non-autoregressive baselines and achieves competitive performance with many strong autoregressive models. In addition, we also conduct extensive analysis experiments to reveal the effect of each proposed component.
Khintchine's theorem is a classical result from metric number theory which relates the Lebesgue measure of certain limsup sets with the convergence/divergence of naturally occurring volume sums. In this paper we ask whether an analogous result holds for iterated function systems (IFSs). We say that an IFS is approximation regular if we observe Khintchine type behaviour, i.e., if the size of certain limsup sets defined using the IFS is determined by the convergence/divergence of naturally occurring sums. We prove that an IFS is approximation regular if it consists of conformal mappings and satisfies the open set condition. The divergence condition we introduce incorporates the inhomogeneity present within the IFS. We demonstrate via an example that such an approach is essential. We also formulate an analogue of the Duffin-Schaeffer conjecture and show that it holds for a set of full Hausdorff dimension. Combining our results with the mass transference principle of Beresnevich and Velani \cite{BerVel}, we prove a general result that implies the existence of exceptional points within the attractor of our IFS. These points are exceptional in the sense that they are "very well approximated". As a corollary of this result, we obtain a general solution to a problem of Mahler, and prove that there are badly approximable numbers that are very well approximated by quadratic irrationals. The ideas put forward in this paper are introduced in the general setting of IFSs that may contain overlaps. We believe that by viewing IFS's from the perspective of metric number theory, one can gain a greater insight into the extent to which they overlap. The results of this paper should be interpreted as a first step in this investigation.
Let (fn)n=1∞ be a sequence of polynomials and α>1. In this paper we study the distribution of the sequence (fn(α))n=1∞ modulo one. We give sufficient conditions for a sequence (fn)n=1∞ to ensure that for Lebesgue almost every α>1 the sequence (fn(α))n=1∞ has Poissonian pair correlations. In particular, this result implies that for Lebesgue almost every α>1, for any k≥2 the sequence (αnk)n=1∞ has Poissonian pair correlations.
Until recently, it was an important open problem in Fractal Geometry to determine whether there exists an iterated function system acting on R with no exact overlaps for which cylinders are super-exponentially close at all small scales. Iterated function systems satisfying these properties were shown to exist by the author and by Bárány and Käenmäki. In this paper we prove a general theorem on the existence of such iterated function systems within a parameterised family. This theorem shows that if a parameterised family contains two independent subfamilies, and the set of parameters that cause exact overlaps satisfies some weak topological assumptions, then the original family will contain an iterated function system satisfying the desired properties. We include several explicit examples of parameterised families to which this theorem can be applied.
We introduce Multi-SimLex, a large-scale lexical resource and evaluation benchmark covering datasets for 12 typologically diverse languages, including major languages (e.g., Mandarin Chinese, Spanish, Russian) as well as less-resourced ones (e.g., Welsh, Kiswahili). Each language dataset is annotated for the lexical relation of semantic similarity and contains 1,888 semantically aligned concept pairs, providing a representative coverage of word classes (nouns, verbs, adjectives, adverbs), frequency ranks, similarity intervals, lexical fields, and concreteness levels. Additionally, owing to the alignment of concepts across languages, we provide a suite of 66 cross-lingual semantic similarity datasets. Due to its extensive size and language coverage, Multi-SimLex provides entirely novel opportunities for experimental evaluation and analysis. On its monolingual and cross-lingual benchmarks, we evaluate and analyze a wide array of recent state-of-the-art monolingual and cross-lingual representation models, including static and contextualized word embeddings (such as fastText, M-BERT and XLM), externally informed lexical representations, as well as fully unsupervised and (weakly) supervised cross-lingual word embeddings. We also present a step-by-step dataset creation protocol for creating consistent, Multi-Simlex-style resources for additional languages. We make these contributions -- the public release of Multi-SimLex datasets, their creation protocol, strong baseline results, and in-depth analyses which can be be helpful in guiding future developments in multilingual lexical semantics and representation learning -- available via a website which will encourage community effort in further expansion of Multi-Simlex to many more languages. Such a large-scale semantic resource could inspire significant further advances in NLP across languages.
A classical theorem of Koksma states that for Lebesgue almost every x>1 the sequence (xn)n=1∞ is uniformly distributed modulo one. In the present paper we extend Koksma's theorem to the pair correlation setting. More precisely, we show that for Lebesgue almost every x>1 the pair correlations of the fractional parts of (xn)n=1∞ are asymptotically Poissonian. The proof is based on a martingale approximation method.
The ability of a dialog system to express prespecified language style during conversations has a direct, positive impact on its usability and on user satisfaction. We introduce a new prototype-to-style (PS) framework to tackle the challenge of stylistic dialogue generation. The framework uses an Information Retrieval (IR) system and extracts a response prototype from the retrieved response. A stylistic response generator then takes the prototype and the desired language style as model input to obtain a high-quality and stylistic response. To effectively train the proposed model, we propose a new style-aware learning objective as well as a de-noising learning strategy. Results on three benchmark datasets from two languages demonstrate that the proposed approach significantly outperforms existing baselines in both in-domain and cross-domain evaluations
In this paper we study the quantitative recurrence properties of self-conformal sets X equipped with the map T:X→X induced by the left shift. In particular, given a function φ:N→(0,∞), we study the metric properties of the set R(T,φ)={x∈X:∣Tnx−x∣<φ(n) for infinitely many n∈N}. Our main result shows that for the natural measure supported on X, R(T,φ) has zero measure if a natural volume sum converges, and under the open set condition R(T,φ) has full measure if this volume sum diverges.
Stylistic response generation is crucial for building an engaging dialogue system for industrial use. While it has attracted much research interest, existing methods often generate stylistic responses at the cost of the content quality (relevance and fluency). To enable better balance between the content quality and the style, we introduce a new training strategy, know as Information-Guided Reinforcement Learning (IG-RL). In IG-RL, a training model is encouraged to explore stylistic expressions while being constrained to maintain its content quality. This is achieved by adopting reinforcement learning strategy with statistical style information guidance for quality-preserving explorations. Experiments on two datasets show that the proposed approach outperforms several strong baselines in terms of the overall response performance.
In this paper we study digit frequencies in the setting of expansions in non-integer bases, and self-affine sets with non-empty interior. Within expansions in non-integer bases we show that if β∈(1,1.787…) then every x∈(0,β−11) has a simply normal β-expansion. We also prove that if β∈(1,21+5) then every x∈(0,β−11) has a β-expansion for which the digit frequency does not exist, and a β-expansion with limiting frequency of zeros p, where p is any real number sufficiently close to 1/2. For a class of planar self-affine sets we show that if the horizontal contraction lies in a certain parameter space and the vertical contractions are sufficiently close to 1, then every nontrivial vertical fibre contains an interval. Our approach lends itself to explicit calculation and give rise to new examples of self-affine sets with non-empty interior. One particular strength of our approach is that it allows for different rates of contraction in the vertical direction.
We study periodic expansions in positional number systems with a base $β\in\C,\ |β|>1$, and with coefficients in a finite set of digits $\A\subset\C.$ We are interested in determining those algebraic bases for which there exists $\A\subset \Q(β),$ such that all elements of $\Q(β)$ admit at least one eventually periodic representation with digits in $\A$. In this paper we prove a general result that guarantees the existence of such an $\A$. This result implies the existence of such an $\A$ when β is a rational number or an algebraic integer with no conjugates of modulus 1. We also consider eventually periodic representations of elements of $\Q(β)$ for which the maximal power of the representation is proportional to the absolute value of the represented number, up to some universal constant. We prove that if every element of $\Q(β)$ admits such a representation then β must be a Pisot number or a Salem number. This result generalises a well known result of Schmidt \cite{Schmidt}.
Given a finite set of real numbers A, the generalised golden ratio is the unique real number G(A)>1 for which we only have trivial unique expansions in smaller bases, and have non-trivial unique expansions in larger bases. We show that G(A) varies continuously with the alphabet A (of fixed size). What is more, we demonstrate that as we vary a single parameter m within~A, the generalised golden ratio function may behave like m1/h for any positive integer h. These results follow from a detailed study of G(A) for ternary alphabets, building upon the work of Komornik, Lai, and Pedicini (2011). We give a new proof of their main result, that is we explicitly calculate the function G({0,1,m}). (For a ternary alphabet, it may be assumed without loss of generality that A={0,1,m} with m∈(1,2)].) We also study the set of m∈(1,2] for which G({0,1,m})=1+m, we prove that this set is uncountable and has Hausdorff dimension~0. We show that the function mapping m to G({0,1,m}) is of bounded variation yet has unbounded derivative. Finally, we show that it is possible to have unique expansions as well as points with precisely two expansions at the generalised golden ratio.
Given β∈(1,2) and x∈[0,β−11], a sequence (εi)i=1∞∈{0,1}N is called a β-expansion for x if x=∑i=1∞βiεi. In a recent article the author studied the quality of approximation provided by the finite sums ∑i=1nεiβ−i \cite{Bak}. In particular, given β∈(1,2) and Ψ:N→R≥0, we associate the set Wβ(Ψ):=⋂m=1∞⋃n=m∞⋃(εi)i=1n∈{0,1}n[∑i=1nβiεi,∑i=1nβiεi+Ψ(n)]. Alternatively, Wβ(Ψ) is the set of x∈R such that for infinitely many n∈N, there exists a sequence (εi)i=1n satisfying the inequalities 0≤x−∑i=1nβiεi≤Ψ(n). If ∑n=1∞2nΨ(n)<∞ then Wβ(Ψ) has zero Lebesgue measure. We call a β∈(1,2) approximation regular, if ∑n=1∞2nΨ(n)=∞ implies Wβ(Ψ) is of full Lebesgue measure within [0,β−11]. The author conjectured in \cite{Bak} that almost every β∈(1,2) is approximation regular. In this paper we make a significant step towards proving this conjecture. The main result of this paper is the following statement: given a sequence of positive real numbers (ωn)n=1∞, which satisfy limn→∞ωn=∞, then for Lebesgue almost every β∈(1.497…,2) the set Wβ(ωn⋅2−n) is of full Lebesgue measure within [0,β−11]. Here the sequence (ωn)n=1∞ should be interpreted as a sequence tending to infinity at a very slow rate.
In this article we study the first return map defined on the switch region induced by the greedy and lazy maps. In particular we study the allowable sequences of return times, and when the first return map is a generalised Lüroth series transformation. We show that there exists a countable collection of disjoint intervals (In)n=1∞, such that all sequences of return times are permissible if and only if β∈In for some n. Moreover, we show that there exists a set M⊆(1,2) of Hausdorff dimension 1 and Lebesgue measure zero, for which the first return map is a generalised Lüroth series transformation if and only if β∈M.
Let q∈(1,2). A q-expansion of a number x in [0,q−11] is a sequence (δi)i=1∞∈{0,1}N satisfying x=∑i=1∞qiδi. Let Bℵ0 denote the set of q for which there exists x with a countable number of q-expansions, and let B1,ℵ0 denote the set of q for which 1 has a countable number of q-expansions. In \cite{Sidorov6} it was shown that minBℵ0=minB1,ℵ0=21+5, and in \cite{Baker} it was shown that Bℵ0∩(21+5,q1]={q1}, where q1(≈1.64541) is the positive root of x6−x4−x3−2x2−x−1=0. In this paper we show that the second smallest point of B1,ℵ0 is q3(≈1.68042), the positive root of x5−x4−x3−x+1=0. Enroute to proving this result we show that Bℵ0∩(q1,q3]={q2,q3}, where q2(≈1.65462) is the positive root of x6−2x4−x3−1=0.
Let β∈(1,2) and x∈[0,β−11]. We call a sequence (εi)i=1∞∈{0,1}N a β-expansion for x if x=∑i=1∞εiβ−i. We call a finite sequence (εi)i=1n∈{0,1}n an n-prefix for x if it can be extended to form a β-expansion of x. In this paper we study how good an approximation is provided by the set of n-prefixes. Given Ψ:N→R≥0, we introduce the following subset of R, Wβ(Ψ):=⋂m=1∞⋃n=m∞⋃(εi)i=1n∈{0,1}n[∑i=1nβiεi,∑i=1nβiεi+Ψ(n)] In other words, Wβ(Ψ) is the set of x∈R for which there exists infinitely many solutions to the inequalities 0≤x−∑i=1nβiεi≤Ψ(n). When ∑n=1∞2nΨ(n)<∞ the Borel-Cantelli lemma tells us that the Lebesgue measure of Wβ(Ψ) is zero. When ∑n=1∞2nΨ(n)=∞, determining the Lebesgue measure of Wβ(Ψ) is less straightforward. Our main result is that whenever β is a Garsia number and ∑n=1∞2nΨ(n)=∞ then Wβ(Ψ) is a set of full measure within [0,β−11]. Our approach makes no assumptions on the monotonicity of Ψ, unlike in classical Diophantine approximation where it is often necessary to assume Ψ is decreasing.
It is a classical result from Diophantine approximation that the set of badly approximable numbers has Lebesgue measure zero. In this paper we generalise this result to more general sequences of balls. Given a countable set of closed d-dimensional Euclidean balls {B(xi,ri)}i=1∞, we say that α∈Rd is a badly approximable number with respect to {B(xi,ri)}i=1∞ if there exists κ(α)>0 and N(α)∈N such that α∈/B(xi,κ(α)ri) for all i≥N(α). Under natural conditions on the set of balls, we prove that the set of badly approximable numbers with respect to {B(xi,ri)}i=1∞ has Lebesgue measure zero. Moreover, our approach yields a new proof that the set of badly approximable numbers has Lebesgue measure zero.
We study the dynamical properties of certain shift spaces. To help study these properties we introduce two new classes of shifts, namely boundedly supermultiplicative (BSM) shifts and balanced shifts. It turns out that any almost specified shift is both BSM and balanced, and any balanced shift is BSM. However, as we will demonstrate, there are examples of shifts which are BSM but not balanced. We also study the measure theoretic properties of balanced shifts. We show that a shift space admits a Gibbs state if and only if it is balanced. Restricting ourselves to S-gap shifts, we relate certain dynamical properties of an S-gap shift to combinatorial properties from expansions in non-integer bases. This identification allows us to use the machinery from expansions in non-integer bases to give straightforward constructions of S-gap shifts with certain desirable properties. We show that for any q∈(0,1) there is an S-gap shift which has the specification property and entropy q. We also use this identification to address the question, for a given q∈(0,1), how many S-gap shifts exist with entropy q? For certain exceptional values of q there is a unique S-gap shift with this entropy.
Let q∈(1,2) and x∈[0,q−11]. We say that a sequence (εi)i=1∞∈{0,1}N is an expansion of x in base q (or a q-expansion) if x=∑i=1∞εiq−i. For any k∈N, let Bk denote the set of q such that there exists x with exactly k expansions in base q. In [12] it was shown that minB2=q2≈1.71064, the appropriate root of x4=2x2+x+1. In this paper we show that for any k≥3, minBk=qf≈1.75488, the appropriate root of x3=2x2−x+1.
Let λ1,…,λn be real numbers in (0,1) and p1,…,pn be points in Rd. Consider the collection of maps fj:Rd→Rd given by fj(x)=λjx+(1−λj)pj. It is a well known result that there exists a unique compact set Λ⊂Rd satisfying Λ=∪j=1nfj(Λ). Each x∈Λ has at least one coding, that is a sequence (εi)i=1∞∈{1,…,n}N that satisfies limN→∞fε1⋯fεN(0)=x. We study the size and complexity of the set of codings of a generic x∈Λ when Λ has positive Lebesgue measure. In particular, we show that under certain natural conditions almost every x∈Λ has a continuum of codings. We also show that almost every x∈Λ has a universal coding. Our work makes no assumptions on the existence of holes in Λ and improves upon existing results when it is assumed Λ contains no holes.
Given a strictly increasing sequence of positive real numbers tending to infinity (qn)n=1∞, and an arbitrary sequence of real numbers (rn)n=1∞. We study the set of α∈(1,∞) for which limn→∞∥αqn−rn∥=0. In \cite{Dub} Dubickas showed that whenever limn→∞(qn+1−qn)=∞, there always exists a transcendental α for which limn→∞∥αqn−rn∥=0. Adapting the approach of Bugeaud and Moshchevitin \cite{BugMos}, we improve upon this result and show that whenever limn→∞(qn+1−qn)=∞, the set of α∈(1,∞) satisfying limn→∞∥αqn−rn∥=0 is a dense set of Hausdorff dimension 1.
Let K⊆R be the unique attractor of an iterated function system. We consider the case where K is an interval and study those elements of K with a unique coding. We prove under mild conditions that the set of points with a unique coding can be identified with a subshift of finite type. As a consequence of this, we can show that the set of points with a unique coding is a graph-directed self-similar set in the sense of Mauldin and Williams \cite{MW}. The theory of Mauldin and Williams then provides a method by which we can explicitly calculate the Hausdorff dimension of this set. Our algorithm can be applied generically, and our result generalises the work of \cite{DKK}, \cite{K1}, \cite{K2}, and \cite{MK}.
Let q∈(1,2) and x∈[0,q−11]. We say that a sequence (εi)i=1∞∈{0,1}N is an expansion of x in base q (or a q-expansion) if x=\sum_{i=1}^{\infty}ε_iq^{-i}. Let Bℵ0 denote the set of q for which there exists x with exactly ℵ0 expansions in base q. In \cite{EHJ} it was shown that minBℵ0=21+5. In this paper we show that the smallest element of Bℵ0 strictly greater than 21+5 is qℵ0≈1.64541, the appropriate root of x6=x4+x3+2x2+x+1. This leads to a full dichotomy for the number of possible q-expansions for q∈(21+5,qℵ0). We also prove some general results regarding Bℵ0∩[21+5,qf], where qf≈1.75488 is the appropriate root of x3=2x2−x+1. Moreover, the techniques developed in this paper imply that if x∈[0,q−11] has uncountably many q-expansions then the set of q-expansions for x has cardinality equal to that of the continuum, this proves that the continuum hypothesis holds when restricted to this specific case.
In this paper we study the topology of a set naturally arising from the study of β-expansions. After proving several elementary results for this set we study the case when our base is Pisot. In this case we give necessary and sufficient conditions for this set to be finite. This finiteness property will allow us to generalise a theorem due to Schmidt and will provide the motivation for sufficient conditions under which the growth rate and Hausdorff dimension of the set of β-expansions are equal and explicitly calculable.
Starting with the work of Lapidus and van Frankenhuysen a number of papers have introduced zeta functions as a way of capturing multifractal information. In this paper we propose a new multifractal zeta function and show that under certain conditions the abscissa of convergence yields the Hausdorff multifractal spectrum for a class of measures.
It is a well known result that for β∈(1,21+5) and x∈(0,β−11) there exists uncountably many (εi)i=1∞∈0,1N such that x=∑i=1∞εiβ−i. When β∈(21+5,2] there exists x∈(0,β−11) for which there exists a unique (εi)i=1∞∈0,1N such that x=∑i=1∞εiβ−i. In this paper we consider the more general case when our sequences are elements of 0,...,mN. We show that an analogue of the golden ratio exists and give an explicit formula for it.
In a recent paper of Feng and Sidorov they show that for β∈(1,21+5) the set of β-expansions grows exponentially for every x∈(0,β−11). In this paper we study this growth rate further. We also consider the set of β-expansions from a dimension theory perspective.
Research graph
Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.
BZPEER is loading the nearby papers, people, topics and institutions for this page.