Machine Learning is the study of methods for programming computers to learn. Computers are applied to a wide range of tasks, and for most of these it is relatively easy for programmers to design and implement the necessary software. However, there are many tasks for which this is difficult or impossible. These can be divided into four general categories. First, there are problems for which there exist no human experts. For example, in modern automated manufacturing facilities, there is a need to predict machine failures before they occur by analyzing sensor readings. Because the machines are new, there are no human experts who can be interviewed by a programmer to provide the knowledge necessary to build a computer system. A machine learning system can study recorded data and subsequent machine failures and learn prediction rules. Second, there are problems where human experts exist, but where they are unable to explain their expertise. This is the case in many perceptual tasks, such as speech recognition, hand-writing recognition, and natural language understanding. Virtually all humans exhibit expert-level abilities on these tasks, but none of them can describe the detailed steps that they follow as they perform them. Fortunately, humans can provide machines with examples of the inputs and correct outputs for these tasks, so machine learning algorithms can learn to map the inputs to the outputs. Third, there are problems where phenomena are changing rapidly. In finance, for example, people would like to predict the future behavior of the stock market, of consumer purchases, or of exchange rates. These behaviors change frequently, so that even if a programmer could construct a good predictive computer program, it would need to be rewritten frequently. A learning program can relieve the programmer of this burden by constantly modifying and tuning a set of learned prediction rules. Fourth, there are applications that need to be customized for each computer user separately. Consider, for example, a program to filter unwanted electronic mail messages. Different users will need different filters. It is unreasonable to expect each user to program his or her own rules, and it is infeasible to provide every user with a software engineer to keep the rules up-to-date. A machine learning system can learn which mail messages the user rejects and maintain the filtering rules automatically. Machine learning addresses many of the same research questions as the fields of statistics, data mining, and psychology, but with differences of emphasis. Statistics focuses on understanding the phenomena that have generated the data, often with the goal of testing different hypotheses about those phenomena. Data mining seeks to find patterns in the data that are understandable by people. Psychological studies of human learning aspire to understand the mechanisms underlying the various learning behaviors exhibited by people (concept learning, skill acquisition, strategy change, etc.).

Machine learning

We propose Prototypical Networks for the problem of few-shot classification, where a classifier must generalize to new classes not seen in the training set, given only a small number of examples of each new class. Prototypical Networks learn a metric space in which classification can be performed by computing distances to prototype representations of each class. Compared to recent approaches for few-shot learning, they reflect a simpler inductive bias that is beneficial in this limited-data regime, and achieve excellent results. We provide an analysis showing that some simple design decisions can yield substantial improvements over recent approaches involving complicated architectural choices and meta-learning. We further extend Prototypical Networks to zero-shot learning and achieve state-of-the-art results on the CU-Birds dataset.

/pdf/prototypical-networks-for-few-shot-learning-2l18ppn4sw.pdf

Prototypical Networks for Few-shot Learning

BIOE 402. Medical Technology Assessment. 2 or 3 hours. Bioentrepreneur course. Assessment of medical technology in the context of commercialization. Objectives, competition, market share, funding, pricing, manufacturing, growth, and intellectual property; many issues unique to biomedical products. Course Information: 2 undergraduate hours. 3 graduate hours. Prerequisite(s): Junior standing or above and consent of the instructor.

“Bioinformatics” 특집을 내면서

We propose a novel formalism, called Frame Logic (abbr., F-logic), that accounts in a clean and declarative fashion for most of the structural aspects of object-oriented and frame-based languages. These features include object identity, complex objects, inheritance, polymorphic types, query methods, encapsulation, and others. In a sense, F-logic stands in the same relationship to the object-oriented paradigm as classical predicate calculus stands to relational programming. F-logic has a model-theoretic semantics and a sound and complete resolution-based proof theory. A small number of fundamental concepts that come from object-oriented programming have direct representation in F-logic; other, secondary aspects of this paradigm are easily modeled as well. The paper also discusses semantic issues pertaining to programming with a deductive object-oriented language based on a subset of F-logic.

/pdf/logical-foundations-of-object-oriented-and-frame-based-2acr2uie5u.pdf

Logical foundations of object-oriented and frame-based languages

Class imbalance is a problem that is common to many application domains. When examples of one class in a training data set vastly outnumber examples of the other class(es), traditional data mining algorithms tend to create suboptimal classification models. Several techniques have been used to alleviate the problem of class imbalance, including data sampling and boosting. In this paper, we present a new hybrid sampling/boosting algorithm, called RUSBoost, for learning from skewed training data. This algorithm provides a simpler and faster alternative to SMOTEBoost, which is another algorithm that combines boosting and data sampling. This paper evaluates the performances of RUSBoost and SMOTEBoost, as well as their individual components (random undersampling, synthetic minority oversampling technique, and AdaBoost). We conduct experiments using 15 data sets from various application domains, four base learners, and four evaluation metrics. RUSBoost and SMOTEBoost both outperform the other procedures, and RUSBoost performs comparably to (and often better than) SMOTEBoost while being a simpler and faster technique. Given these experimental results, we highly recommend RUSBoost as an attractive alternative for improving the classification performance of learners built using imbalanced data.

/pdf/rusboost-a-hybrid-approach-to-alleviating-class-imbalance-4wo2psaj6r.pdf

RUSBoost: A Hybrid Approach to Alleviating Class Imbalance

Object-oriented databases have been introduced primarily to ease the development of database applications. However, the difficulties encountered when, for instance, trying to restructure data or integrate databases demonstrate that the models being used still lack flexibility. We claim that the natural way to overcome these shortcomings is to introduce a sophisticated view mechanism. This paper presents such a mechanism, one which allows a programmer to restructure the class hierarchy and modify the behavior and structure of objects. The mechanism allows a programmer to specify attribute values implicitly, rather than storing them. It also allows him to introduce new classes into the class hierarchy. These virtual classes are populated by selecting existing objects from other classes and by creating new objects. Fixing the identify of new objects during database updates introduces subtle issues into view design. Our presentation, mostly informal, leans on a number of illustrative examples meant to emphasize the simplicity of our mechanism.

Objects and views

Seed germination is a complex trait of key ecological and agronomic significance. Few genetic factors regulating germination have been identified, and the means by which their concerted action controls this developmental process remains largely unknown. Using publicly available gene expression data from Arabidopsis thaliana, we generated a condition-dependent network model of global transcriptional interactions (SeedNet) that shows evidence of evolutionary conservation in flowering plants. The topology of the SeedNet graph reflects the biological process, including two state-dependent sets of interactions associated with dormancy or germination. SeedNet highlights interactions between known regulators of this process and predicts the germination-associated function of uncharacterized hub nodes connected to them with 50% accuracy. An intermediate transition region between the dormancy and germination subdomains is enriched with genes involved in cellular phase transitions. The phase transition regulators SERRATE and EARLY FLOWERING IN SHORT DAYS from this region affect seed germination, indicating that conserved mechanisms control transitions in cell identity in plants. The SeedNet dormancy region is strongly associated with vegetative abiotic stress response genes. These data suggest that seed dormancy, an adaptive trait that arose evolutionarily late, evolved by coopting existing genetic pathways regulating cellular phase transition and abiotic stress. SeedNet is available as a community resource (http://vseed.nottingham.ac.uk) to aid dissection of this complex trait and gene function in diverse processes.

https://www.pnas.org/content/pnas/108/23/9709.full.pdf

Genome-wide network model capturing seed germination reveals coordinated regulation of plant cellular phase transitions.

This paper presents an overview of Transaction Logic —a new formalism recently introduced in Bonner and Kifer (1992, 1993) and designed to deal with the phenomenon of state changes in logic programming, databases, and AI. Transaction Logic has a natural model theory and a sound and complete proof theory. Unlike many other logics, however, it is suitable for programming procedures that accomplish state transitions in a logically sound manner. Transaction logic amalgamates such features as hypothetical and committed updates, dynamic constraints on transaction execution, nondeterminism, and bulk updates. Transaction Logic also appears to be suitable as a logical model of hitherto nonlogical phenomena, including so-called procedural knowledge in AI, and the behavior of object-oriented databases, especially methods with side effects.

An overview of transaction logic

An extension of predicate logic, called Transaction Logic, is proposed, which accounts in a clean and declarative fashion for the phenomenon of state changes in logic programs and databases. Transaction Logic has a natural model theory and a sound and complete proof theory, but unlike many other logics, it allows users to program transactions. The semantics leads naturally to features whose amalgamation in a single logic has proved elusive in the past. These features include both hypothetical and committed updates, dynamic constraints on transaction execution, non-determinism, and bulk updates. Finally, Transaction Logic holds promise as a logical model of hitherto non-logical phenomena, including so-called procedural knowledge in AI, and the behavior of object-oriented databases, especially methods with side eeects. This paper presents the semantics of Transaction Logic coupled with a sound and complete proof theory for a Horn-like subset of the logic.

Transaction logic programming

In previous work, we developed Transaction Logic (or T R), which deals with state changes in deductive databases. T R provides a logical framework in which elementary database updates and queries can be combined into complex database transactions. T R accounts not only for the updates themselves, but also for important related problems, such as the order of update operations, non-determinism, and transaction failure and rollback. In the present paper, we propose Concurrent Transaction Logic (or CT R), which extends Transaction Logic with connectives for modeling the concurrent execution of complex processes. Concurrent processes in CT R execute in an interleaved fashion and can communicate and synchronize themselves. Like classical logic, CT R has a \Horn" fragment that has both a procedural and a declarative semantics, in which users can program and execute database transactions. CT R is thus a deductive database language that integrates concurrency, communication, and updates. All this is accomplished in a completely logical framework, including a natural model theory and a proof theory. Moreover, this framework is exible enough to accommodate many di erent semantics for updates and deductive databases. For instance, not only can updates insert and delete tuples, they can also insert and delete null values, rules, or arbitrary logical formulas. Likewise, not only can databases have a classical semantics, they can also have the well-founded semantics, the stable-model semantics, etc. Finally, the proof theory for CT R has an e cient SLD-style proof procedure. As in the sequential version of the logic, this proof procedure not only nds proofs, it also executes concurrent transactions, nds their execution schedules, and updates the database. A main result is that the proof theory is sound and complete for the model theory. Appears in Proceedings of the Joint International Conference and Symposium on Logic Programming (JICSLP), September 2{6 1996 Bonn, Germany, pages 142{156. MIT Press. This and related papers and software are available at the following web page: http://www.cs.toronto.edu/~bonner/transaction-logic.html Work supported in part by an Operating Grant from the Natural Sciences and Engineering Research Council of Canada. Supported in part by the NSF Grant IRI-9404629.

Anthony J. Bonner

Papers

Objects and views

Genome-wide network model capturing seed germination reveals coordinated regulation of plant cellular phase transitions.

An overview of transaction logic

Transaction logic programming

Concurrency and Communication in Transaction Logic.