Motivated in part by their relevance for low-precision training and quantization, massive activations in large language models (LLMs) have recently emerged as a topic of interest. However, existing analyses are limited in scope, and generalizability across architectures is unclear. This paper helps address some of these gaps by conducting an analysis of massive activations across a broad range of LLMs, including both GLU-based and non-GLU-based architectures. Our findings challenge several prior assumptions, most importantly: (1) not all massive activations are detrimental, i.e. suppressing them does not lead to an explosion of perplexity or a collapse in downstream task performance; (2) proposed mitigation strategies such as Attention KV bias are model-specific and ineffective in certain cases. We consequently investigate novel hybrid mitigation strategies; in particular pairing Target Variance Rescaling (TVR) with Attention KV bias or Dynamic Tanh (DyT) successfully balances the mitigation of massive activations with preserved downstream model performance in the scenarios we investigated. Our code is available at: https://github.com/bluorion-com/refine_massive_activations.

The outcome of Large Language Model (LLM) pre-training strongly depends on weight initialization and variance control strategies. Although the importance of initial variance control has been well documented in neural networks in general, the literature on initialization and management of its growth during LLM pre-training, specifically, is somewhat sparse. In this paper, we introduce the Layer Index Rescaling (LIR) weight initialization scheme, and the Target Variance Rescaling (TVR) variance control strategy. Experiments on a 1B parameter LLaMA model demonstrate that better variance management using these techniques yields substantial improvements in downstream task performance (up to 4.6% on common pre-training benchmarks) and reduces extreme activation values, thus mitigating challenges associated with quantization and low-precision training.

The recent breakthroughs in Large Language Models (LLMs) have mostly focused on languages with easily available and sufficient resources, such as English. However, there remains a significant gap for languages that lack sufficient linguistic resources in the public domain. Our work introduces Komodo-7B, 7-billion-parameter Large Language Models designed to address this gap by seamlessly operating across Indonesian, English, and 11 regional languages in Indonesia. Komodo-7B is a family of LLMs that consist of Komodo-7B-Base and Komodo-7B-Instruct. Komodo-7B-Instruct stands out by achieving state-of-the-art performance in various tasks and languages, outperforming the benchmarks set by OpenAI's GPT-3.5, Cohere's Aya-101, Llama-2-Chat-13B, Mixtral-8x7B-Instruct-v0.1, Gemma-7B-it , and many more. This model not only demonstrates superior performance in both language-specific and overall assessments but also highlights its capability to excel in linguistic diversity. Our commitment to advancing language models extends beyond well-resourced languages, aiming to bridge the gap for those with limited linguistic assets. Additionally, Komodo-7B-Instruct's better cross-language understanding contributes to addressing educational disparities in Indonesia, offering direct translations from English to 11 regional languages, a significant improvement compared to existing language translation services. Komodo-7B represents a crucial step towards inclusivity and effectiveness in language models, providing to the linguistic needs of diverse communities.

This paper is a contribution to the reproducibility challenge in the field of machine learning, specifically addressing the issue of certifying the robustness of neural networks (NNs) against adversarial perturbations. The proposed Double Sampling Randomized Smoothing (DSRS) framework overcomes the limitations of existing methods by using an additional smoothing distribution to improve the robustness certification. The paper provides a clear manifestation of DSRS for a generalized family of Gaussian smoothing and a computationally efficient method for implementation. The experiments on MNIST and CIFAR-10 demonstrate the effectiveness of DSRS, consistently certifying larger robust radii compared to other methods. Also various ablations studies are conducted to further analyze the hyperparameters and effect of adversarial training methods on the certified radius by the proposed framework.

This paper introduces a novel method leveraging bi-encoder-based detectors along with a comprehensive study comparing different out-of-distribution (OOD) detection methods in NLP using different feature extractors. The feature extraction stage employs popular methods such as Universal Sentence Encoder (USE), BERT, MPNET, and GLOVE to extract informative representations from textual data. The evaluation is conducted on several datasets, including CLINC150, ROSTD-Coarse, SNIPS, and YELLOW. Performance is assessed using metrics such as F1-Score, MCC, FPR@90, FPR@95, AUPR, an AUROC. The experimental results demonstrate that the proposed bi-encoder-based detectors outperform other methods, both those that require OOD labels in training and those that do not, across all datasets, showing great potential for OOD detection in NLP. The simplicity of the training process and the superior detection performance make them applicable to real-world scenarios. The presented methods and benchmarking metrics serve as a valuable resource for future research in OOD detection, enabling further advancements in this field. The code and implementation details can be found on our GitHub repository: https://github.com/yellowmessenger/ood-detection.

We propose a Multi-Task Learning (MTL) paradigm based deep neural network architecture, called MTCNet (Multi-Task Crowd Network) for crowd density and count estimation. Crowd count estimation is challenging due to the non-uniform scale variations and the arbitrary perspective of an individual image. The proposed model has two related tasks, with Crowd Density Estimation as the main task and Crowd-Count Group Classification as the auxiliary task. The auxiliary task helps in capturing the relevant scale-related information to improve the performance of the main task. The main task model comprises two blocks: VGG-16 front-end for feature extraction and a dilated Convolutional Neural Network for density map generation. The auxiliary task model shares the same front-end as the main task, followed by a CNN classifier. Our proposed network achieves 5.8% and 14.9% lower Mean Absolute Error (MAE) than the state-of-the-art methods on ShanghaiTech dataset without using any data augmentation. Our model also outperforms with 10.5% lower MAE on UCF_CC_50 dataset.

This paper proposes a Convolutional Neural Network (CNN) inspired by Multitask Learning (MTL) and based on speech features trained under the joint supervision of softmax loss and center loss, a powerful metric learning strategy, for the recognition of emotion in speech. Speech features such as Spectrograms and Mel-frequency Cepstral Coefficient s (MFCCs) help retain emotion-related low-level characteristics in speech. We experimented with several Deep Neural Network (DNN) architectures that take in speech features as input and trained them under both softmax and center loss, which resulted in highly discriminative features ideal for Speech Emotion Recognition (SER). Our networks also employ a regularizing effect by simultaneously performing the auxiliary task of reconstructing the input speech features. This sharing of representations among related tasks enables our network to better generalize the original task of SER. Some of our proposed networks contain far fewer parameters when compared to state-of-the-art architectures.

We propose a novel visual context-aware filter generation module which incorporates contextual information present in images into Convolutional Neural Networks (CNNs). In contrast to traditional CNNs, we do not employ the same set of learned convolution filters for all input image instances. Our proposed input-conditioned convolution filters when combined with techniques inspired by Multi-instance learning and max-pooling, results in a transformation-invariant neural network. We investigated the performance of our proposed framework on three MNIST variations, which covers both rotation and scaling variance, and achieved 1.13% error on MNIST-rot-12k, 1.12% error on Half-rotated MNIST and 0.68% error on Scaling MNIST, which is significantly better than the state-of-the-art results. We make use of visualization to further prove the effectiveness of our visual context-aware convolution filters. Our proposed visual context-aware convolution filter generation framework can also serve as a plugin for any CNN based architecture and enhance its modeling capacity.

This paper proposes a speech emotion recognition method based on speech features and speech transcriptions (text). Speech features such as Spectrogram and Mel-frequency Cepstral Coefficients (MFCC) help retain emotionrelated low-level characteristics in speech whereas text helps capture semantic meaning, both of which help in different aspects of emotion detection. We experimented with several Deep Neural Network (DNN) architectures, which take in different combinations of speech features and text as inputs. The proposed network architectures achieve higher accuracies when compared to state-of-the-art methods on a benchmark dataset. The combined MFCC-Text Convolutional Neural Network (CNN) model proved to be the most accurate in recognizing emotions in IEMOCAP data. We achieved an almost 7% increase in overall accuracy as well as an improvement of 5.6% in average class accuracy when compared to existing state-of-the-art methods.

This paper proposes a Residual Convolutional Neural Network (ResNet) based on speech features and trained under Focal Loss to recognize emotion in speech. Speech features such as Spectrogram and Mel-frequency Cepstral Coefficients (MFCCs) have shown the ability to characterize emotion better than just plain text. Further Focal Loss, first used in One-Stage Object Detectors, has shown the ability to focus the training process more towards hard-examples and down-weight the loss assigned to well-classified examples, thus preventing the model from being overwhelmed by easily classifiable examples. After experimenting with several Deep Neural Network (DNN) architectures, we propose a ResNet, which takes in Spectrogram or MFCC as input and supervised by Focal Loss, ideal for speech inputs where there exists a large class imbalance. Maintaining continuity with previous work in this area, we have used the University of Southern California's Interactive Emotional Motion Capture (USC-IEMOCAP) database's Improvised Topics in this work. This dataset is ideal for our work, as there exists a significant class imbalance among the various emotions. Our best model achieved a 3.4% improvement in overall accuracy and a 2.8% improvement in class accuracy when compared to existing state-of-the-art methods.