Skeleton-based two-person interaction recognition has been gaining increasing attention as advancements are made in pose estimation and graph convolutional networks. Although the accuracy has been gradually improving, the increasing computational complexity makes it more impractical for a real-world environment. There is still room for accuracy improvement as the conventional methods do not fully represent the relationship between inter-body joints. In this paper, we propose a lightweight model for accurately recognizing two-person interactions. In addition to the architecture, which incorporates middle fusion, we introduce a factorized convolution technique to reduce the weight parameters of the model. We also introduce a network stream that accounts for relative distance changes between inter-body joints to improve accuracy. Experiments using two large-scale datasets, NTU RGB+D 60 and 120, show that our method simultaneously achieved the highest accuracy and relatively low computational complexity compared with the conventional methods.

We study the problem of finding small universal reversible Turing machines (URTMs) with four symbols or less. Here, we present two models of URTMs: a 24-state 4-symbol URTM, and a 32-state 3-symbol URTM. Both of them are designed so that they can simulate cyclic tag systems, a kind of string rewriting systems that are universal. By converting the 24-state 4-symbol URTM into a 2-symbol machine, we also obtain a 138-state 2-symbol URTM.

Random fluctuations will be a major factor interfering with the operation of nanometer scale electronic devices. This article presents circuit architectures that can exploit such fluctuations, if signals have a particle-like (discrete, token-based) character. We define an abstract circuit primitive that, though lacking functionality when used with fluctuation-free signals, becomes universal when fluctuations are allowed. Key to the power of a signal’s fluctuations is the ability to explore the state space of a circuit. This ability is used to resolve deadlock situations, which could otherwise only be averted by increased design complexity. The results in this article suggest that in the design of future computers, signal fluctuations, rather than being an impediment to be avoided at any cost, may be an important ingredient to achieve efficient operation.

A two-way reversible multi-head finite automaton (RMFA) is introduced as a simple model of reversible computing, and its language accepting capability is studied. We show that various non-regular context-free languages, and non-context-free context-sensitive languages are accepted by RMFAs with a few heads. For example, we give an RMFA with three heads that accepts the language consisting of all words whose length is a prime number. A construction method of a garbage-less RMFA from a given RFMA is also shown.

Delay-insensitive (DI) circuits are a class of asynchronous circuits whose correctness of operation is robust to arbitrary delays in modules or interconnection lines. Keller clarified the precise operating conditions of the class of DI-circuits and presented a universal set of primitive modules from which any circuit in the class is realizable. Later, Patra and Fussell presented an alternative universal set of primitive modules and claimed that there is no universal set of primitives satisfying Keller's conditions in which the largest number of input and output lines of each primitive module is less than five. We present new types of primitive modules, each having at most three input and output-lines and show they form a universal set of primitives. We achieve this reduction in complexity by allowing the input and output-lines of modules to be bidirectional and to be able to buffer signals. The use of buffers in interconnection lines allows higher throughput of signals and results in circuits requiring less feedback lines, thus improving the efficiency of DI-circuits. The proposed class of Dl-circuits is especially useful for implementations on cellular automata - an architecture that promises efficient implementations and manufacturing in nanotechnology due to its regular structure.

Due to inevitable power dissipation, it is said that nano-scaled computing devices should perform their computing processes in a reversible manner. This will be a large problem in constructing three-dimensional nano-scaled functional objects. Reversible cellular automata (RCA) are used for modeling physical phenomena such as power dissipation, by studying the dissipation of garbage signals. We construct a three-dimensional self-inspective self-reproducing reversible cellular automaton by extending the two-dimensional version SR(8). It can self-reproduce various patterns in three-dimensional reversible cellular space without dissipating garbage signals.

A PCAAG introduced by Morita and Ueno is a parallel array generator on a partitioned cellular automaton (PCA) that generates an array language (i.e. a set of symbol arrays). A "reversible" PCAAG (RPCAAG) is a backward deterministic PCAAG, and thus parsing of two-dimensional patterns can be performed without backtracking by an "inverse" system of the RPCAAG. Hence, a parallel pattern recognition mechanism on a deterministic cellular automaton can be directly obtained from a RPCAAG that generates the pattern set. In this paper, we investigate the generating ability of RPCAAGs and their subclass. It is shown that the ability of RPCAAGs is characterized by two-dimensional deterministic Turing machines, i.e. they are universal in their generating ability. We then investigate a monotonic RPCAAG (MRPCAAG), which is a special type of an RPCAAG that satisfies monotonic constraint. We show that the generating ability of MRPCAAGs is exactly characterized by two-dimensional deterministic linear-bounded automata.

We propose a new system of generating array languages in parallel, based on a partitioned cellular automaton (PCA), a kind of cellular automaton. This system is called a PCA array generator (PCAAG). The characteristic of PCAAG is that a”reversible” version is easily defined. A reversible PCA (RPCA) is a backward deterministic PCA, and we can construct a deterministic “inverse” PCA that undoes the operations of the RPCA. Thus if an array language is generated by an RPCA, it can be parsed in parallel by a deterministic inverse PCA without backtracking. We also define two subclasses of PCAAG, and give examples of them that generate geometrical figures.

We propose a new subclass of two-dimensional isometric array grammar, called Uniquely Parsable Array Grammar (UPAG). UPAG is a grammar class such that when we parse (i.e. apply rewriting rules reversely to) a given two-dimensional word, the rewritable portions of the word do not overlap each other and the rule that is applicable to each portion is uniquely determined. Thus parsing can be done in a deterministic manner. We study basic properties of UPAG such as unique parsability and parallel parsability. Then we define Monotone Terminating UPAG (MTUPAG), a subclass of UPAG, and show that any two-dimensional word generated by MTUPAG can be parsed in linear time. We also show several examples of MTUPAGs which generate interesting sets of geometrical patterns.

Two kinds of three-way isometric array grammars arc proposed as subclasses of an isometric monotonic array grammar. They are a three-way horizontally context-sensitive array grammar (3HCSAG) and a three-way immediately terminating array grammar (3ITAG). In these three-way grammars, patterns of symbols can grow only in the leftward, rightward and downward directions. We show that their generating abilities of rectangular languages are precisely characterized by some kinds of three-way two-dimensional Turing machines or related acceptors. In this paper. the following results are proved. First, 3HCSAG is characterized by a nondeterministic two-dimensional three-way Turing machine with space-bound n (n is the width of a rectangular input) and a nondeterministic one-way parallel/sequential array acceptor. Second, 3ITAG is characterized by a nondeterministic two-dimensional three-way real-time (or linear-time) restricted Turing machine, a nondeterministic one-dimensional bounded cellular acceptor and a nondeterministic two-dimensional one-line tessellation acceptor.

Regular array grammars (RAGs) are the lowest subclass in the Chomsky-like hierarchy of isometric array grammars. The left-hand side of each rewriting rule of RAGs has one nonterminal symbol and at most one "#" (a blank symbol). Therefore, the rewriting rules cannot sense contexts of non-# symbols. However, they can sense # as a kind of context. In this paper, we investigate this #-sensing ability. and study the language generating power of RAGs. Making good use of this ability, We show a method for RAGs to sense the contexts of local shapes of a host array in a derivation. Using this method, we give RAGs which generate the sets of all solid upright rectangles and all solid squares. On the other hand. it is proved that there is no context-free array grammar (and thus no RAG) which generates the set of all hollow upright rectangles.