Natural Language Generation (NLG) typically involves evaluating the generated text in various aspects (e.g., consistency and naturalness) to obtain a comprehensive assessment. However, multi-aspect evaluation remains challenging as it may require the evaluator to generalize to any given evaluation aspect even if it's absent during training. In this paper, we introduce X-Eval, a two-stage instruction tuning framework to evaluate the text in both seen and unseen aspects customized by end users. X-Eval consists of two learning stages: the vanilla instruction tuning stage that improves the model's ability to follow evaluation instructions, and an enhanced instruction tuning stage that exploits the connections between fine-grained evaluation aspects to better assess text quality. To support the training of X-Eval, we collect AspectInstruct, the first instruction tuning dataset tailored for multi-aspect NLG evaluation spanning 27 diverse evaluation aspects with 65 tasks. To enhance task diversity, we devise an augmentation strategy that converts human rating annotations into diverse forms of NLG evaluation tasks, including scoring, comparison, ranking, and Boolean question answering. Extensive experiments across three essential categories of NLG tasks: dialogue generation, summarization, and data-to-text coupled with 21 aspects in meta-evaluation, demonstrate that our X-Eval enables even a lightweight language model to achieve a comparable if not higher correlation with human judgments compared to the state-of-the-art NLG evaluators, such as GPT-4.

We propose CHRT (Control Hidden Representation Transformation) - a controlled language generation framework that steers large language models to generate text pertaining to certain attributes (such as toxicity). CHRT gains attribute control by modifying the hidden representation of the base model through learned transformations. We employ a contrastive-learning framework to learn these transformations that can be combined to gain multi-attribute control. The effectiveness of CHRT is experimentally shown by comparing it with seven baselines over three attributes. CHRT outperforms all the baselines in the task of detoxification, positive sentiment steering, and text simplification while minimizing the loss in linguistic qualities. Further, our approach has the lowest inference latency of only 0.01 seconds more than the base model, making it the most suitable for high-performance production environments. We open-source our code and release two novel datasets to further propel controlled language generation research.

In this paper, we tackle the issue of timbre transfer on a given monophonic music sample. The objective is to change the timbre of source audio from one instrument to another while preserving features such as loudness, pitch, and rhythm. Existing approaches use image-to-image translation techniques on the entire region of time-frequency representations of the raw audio wave, which may lead to the addition of unwanted elements in the final audio waveform. We propose Attention-based Timbre Transfer (ATT), an attention-based pipeline for transferring timbre. To the best of our knowledge, ATT is the first approach which leverages attention for achieving timbre transfer. Further, ATT uses MelGAN for spectrogram inversion, which provides a fast and parallel alternative to other autoregressive music generation approaches, without compromising on the quality. ATT shows promising results, thus efficaciously transferring timbre with minimal offset to other physical characteristics.

Word embeddings are the standard model for semanticandsyntactic representations ofwords. Unfortunately, these models have been shown to exhibit undesirableword associations resulting from gender, racial, and religious biases. Existing post-processing methods for debiasing word embeddings are unable to mitigate gender bias hidden in the spatial arrangement of word vectors. In this paper, we propose RAN-Debias, a novel gender debiasing methodology that not only eliminates the bias present in a word vector but also alters the spatial distribution of its neighboring vectors, achieving a bias- free setting whilemaintaining minimal semantic offset. We also propose a new bias evaluation metric, Gender-based Illicit Proximity Estimate (GIPE), which measures the extent of undue proximity in word vectors resulting from the presence of gender-based predilections. Experiments based on a suite of evaluation metrics show that RAN- Debias significantly outperforms the state-of-the-art in reducing proximity bias (GIPE) by at least 42.02%. It also reduces direct bias, adding minimal semantic disturbance, and achieves the best performance in a downstream application task (coreference resolution).

With the rapid increase of internet users all over the world, social media platforms have risen at a tremendous pace. Code-switched languages (When the speaker alternates between two or more languages eg. Hinglish, Hindi words written in English) are a popular medium of communication on social media. They are characterized by the lack of grammatical structure and variation in spellings. These linguistic constraints combined with lack of data cause ambiguity making the task of text classification on code-switched data difficult. In this paper, we have proposed a Language Modelling (LM) based approach to text classification of Hinglish text. We approach this problem by building a Universal Language Model Fine-tuning using AWD-LSTM architecture on a Hindi-English code-switched (Hinglish) corpus collected from various blogging sites. The language model is able to encode important information about the code-switched data and can be quickly fine-tuned on a given Hinglish dataset and achieve good results. We evaluated the performance of our model on the code-switched aggression detection TRAC-1 dataset, Hinglish Offensive Tweet (HOT) dataset and humour-classification dataset. Experiments on these datasets using our proposed method were able to surpass the previously reported results.

Present-day organizations continue to expose their critical information infrastructures over the Internet for facilitating accessibility; substantially raising concerns about the security of data from both outsiders and insiders. In this paper, we propose a novel approach for detecting intrusive attacks on databases by assessing the risk for incoming transaction based upon the conflation of multiple behavior-based components for the user. In a database intrusion detection system for a role-based access (RBAC) environment, it is not sufficient to focus on role-based features as every user within the same role has a degree of uniqueness. Moreover, traditional database intrusion detection systems classify the incoming transactions into two classes (Malicious or Non-malicious), taking the same action for all transactions that are labeled as malicious irrespective of the damage it can cause to the system. Our approach, Role and User Behavior-based Risk Assessment (RUBRA) uses both role-behavior and user-behavior based features for detecting an intrusive attack. Further, we also quantify the risk associated with the incoming transaction, streamlining the countermeasure process. Experiments on stochastic datasets show promising results on both detection and labeling of malicious transactions.

The area of Handwritten Signature Verification has been broadly researched in the last decades, but remains an open research problem. In offline (static) signature verification, the dynamic information of the signature writing process is lost, and it is difficult to design good feature extractors that can distinguish genuine signatures and skilled forgeries. This verification task is even harder in writer independent scenarios which is undeniably fiscal for realistic cases. In this paper, we have proposed an Ensemble model for offline writer, independent signature verification task with Deep learning. We have used two CNNs for feature extraction, after that RGBT for classification & Stacking to generate final prediction vector. We have done extensive experiments on various datasets from various sources to maintain a variance in the dataset. We have achieved the state of the art performance on various datasets.

Resource management and job scheduling are two problems that go hand-in-hand and the solutions to which are primarily dependent on the nature of workload. With increasing demand to automate the entire process from allocating resources to scheduling jobs efficiently, deep reinforcement learning techniques have been brought into the picture which adapt to the environment and learn from experience. In this paper, we present SchedQRM which classifies burst time of jobs based on their signature and employs Deep Q-Network algorithm to find an optimal solution for any arbitrary job set. We also evaluate our proposed work against state-of-the-art heuristics to show the efficacy of our approach. Keywords— reinforcement learning, job scheduling , Deep-Q Network

Code-mixing is the phenomenon of mixing the vocabulary and syntax of multiple languages in the same sentence. It is an increasingly common occurrence in today's multilingual society and poses a big challenge when encountered in different downstream tasks. In this paper, we present a hybrid architecture for the task of Sentiment Analysis of English-Hindi code-mixed data. Our method consists of three components, each seeking to alleviate different issues. We first generate sub-word level representations for the sentences using a CNN architecture. The generated representations are used as inputs to a Dual Encoder Network which consists of two different BiLSTMs - the Collective and Specific Encoder. The Collective Encoder captures the overall sentiment of the sentence, while the Specific Encoder utilizes an attention mechanism in order to focus on individual sentiment-bearing sub-words. This, combined with a Feature Network consisting of orthographic features and specially trained word embeddings, achieves state-of-the-art results 83.54% accuracy and 0.827 F1 score - on a benchmark dataset.