International Business Machines Corporation

George Kour

Boaz Carmeli

Guy Hadash

Ofer Lavi

Naama Tepper

Ateret Anaby Tavor

Esther Goldbraich

Naama Zwerdling

Amir Kantor

Segev Shlomov

Guy Lev

Einat Kermany

Yftah Ziser

Methods and systems for a reinforcement learning system. A spatial and temporal representation of an observed state of an environment is encoded. A previous state is estimated from a given state and a size of a reward is adjusted based on a difference between the estimated previous state and the previous state.

Removing unnecessary history from reinforcement learning state

Balancing an imbalanced dataset, by: Receiving a balancing policy and the imbalanced dataset. Performing initial adjustment of the imbalanced dataset to comply with the balancing policy, by: oversampling one or more underrepresented classes, and, if one or more of the classes are overrepresented, undersampling them. Operating a generative machine learning model to generate samples for the one or more underrepresented classes, based on the initially-adjusted dataset. Operating a machine learning classification model to label the generated samples with class labels corresponding to the one or more underrepresented classes. Selecting some of the generated samples which, according to the labeling, have a relatively high probability of preserving their class labels. Composing a balanced dataset which complies with the balancing policy and comprises: the samples belonging to the one or more underrepresented classes, the selected generated samples, and an undersampling of the samples belonging to the one or more overrepresented classes.

Dataset balancing via quality-controlled sample generation

A specification of a property required to be upheld by a computerized machine learning system is obtained. A training data set corresponding to the property and inputs and outputs of the system is built. The system is trained on the training data set. Activity of the system is monitored before, during, and after the training. Based on the monitoring, performance of the system is evaluated to determine whether the system, once trained on the training data set, upholds the property.

Deep learning testing

Embodiments of the present systems and methods may provide techniques for augmenting textual data that may be used for textual classification tasks. Embodiments of such techniques may provide the capability to synthesize labeled data to improve text classification tasks. Embodiments may be specifically useful when only a small amount of data is available, and provide improved performance in such cases. For example, in an embodiment, a method implemented in a computer system may comprise a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor, and the method may comprise fine-tuning a language model using a training dataset, synthesizing a plurality of samples using the fine-tuned language model, filtering the plurality of synthesized samples, and generating an augmented training dataset comprising the training dataset and the filtered plurality of synthesized sentences.

Language-model-based data augmentation method for textual classification tasks with little data

In some examples, a system for executing natural language processing techniques can include a processor to detect text comprising a word and a number. The processor can also embed, via a word embedding model, the word into a first vector of a vector space and embed the number by converting the number into a second vector of the vector space. Additionally, the processor can train a deep neural network to execute instructions based on the first embedded vector of the word and the second embedded vector of the number. Furthermore, the processor can process an instruction based on the trained deep neural network.

Growing community of inventors

Tel Aviv, Israel

George Kour