Robust Adversarial Imitation Learning via Adaptively-Selected Demonstrations

The agent in imitation learning (IL) is expected to mimic the behavior of the expert. Its performance relies highly on the quality of given expert demonstrations. However, the assumption that collected demonstrations are optimal cannot always hold in real-world tasks, which would seriously influence the performance of the learned agent. In this paper, we propose a robust method within the framework of Generative Adversarial Imitation Learning (GAIL) to address imperfect demonstration issue, in which good demonstrations can be adaptively selected for training while bad demonstrations are abandoned. Specifically, a binary weight is assigned to each expert demonstration to indicate whether to select it for training. The reward function in GAIL is employed to determine this weight (i.e. higher reward results in higher weight). Compared to some existing solutions that require some auxiliary information about this weight, we set up the connection between weight and model so that we can jointly optimize GAIL and learn the latent weight. Besides hard binary weighting, we also propose a soft weighting scheme. Experiments in the Mujoco demonstrate the proposed method outperforms other GAIL-based methods when dealing with imperfect demonstrations.

An Efficient Method for Video Shot Transition Detection Using Probability Binary Weight Approach

Shot boundary detection in videos is one of the most fundamental tasks towards content-based video retrieval and analysis. In this aspect, an efficient approach to detect abrupt and gradual transition in videos is presented. The proposed method detects the shot boundaries in videos by extracting block-based mean probability binary weight (MPBW) histogram from the normalized Kirsch magnitude frames as an amalgamation of local and global features. Abrupt transitions in videos are detected by utilizing the distance measure between consecutive MPBW histograms and employing an adaptive threshold. In the subsequent step, co-efficient of mean deviation and variance statistical measure is applied on MPBW histograms to detect gradual transitions in the video. Experiments were conducted on TRECVID 2001 and 2007 datasets to analyse and validate the proposed method. Experimental result shows significant improvement of the proposed SBD approach over some of the state-of-the-art algorithms in terms of recall, precision, and F1-score.

Mixed-precision weights network for field-programmable gate array

In this study, we introduced a mixed-precision weights network (MPWN), which is a quantization neural network that jointly utilizes three different weight spaces: binary {−1,1}, ternary {−1,0,1}, and 32-bit floating-point. We further developed the MPWN from both software and hardware aspects. From the software aspect, we evaluated the MPWN on the Fashion-MNIST and CIFAR10 datasets. We systematized the accuracy sparsity bit score, which is a linear combination of accuracy, sparsity, and number of bits. This score allows Bayesian optimization to be used efficiently to search for MPWN weight space combinations. From the hardware aspect, we proposed XOR signed-bits to explore floating-point and binary weight spaces in the MPWN. XOR signed-bits is an efficient implementation equivalent to multiplication of floating-point and binary weight spaces. Using the concept from XOR signed bits, we also provide a ternary bitwise operation that is an efficient implementation equivalent to the multiplication of floating-point and ternary weight space. To demonstrate the compatibility of the MPWN with hardware implementation, we synthesized and implemented the MPWN in a field-programmable gate array using high-level synthesis. Our proposed MPWN implementation utilized up to 1.68-4.89 times less hardware resources depending on the type of resources than a conventional 32-bit floating-point model. In addition, our implementation reduced the latency up to 31.55 times compared to 32-bit floating-point model without optimizations.

Modification of systemic anti-cancer therapies and weight loss, a population-level real-world evidence study

Background: Involuntary weight loss may occur during systemic anti-cancer therapy (SACT), causing treatment disruption and poorer prognoses. There remain gaps in clinical awareness as to which patients may benefit from nutritional interventions that aim to prevent unintended weight loss during SACT. We utilised England’s population-level cancer registry data, conducting a pan-cancer assessment of patient weight loss during SACT. We aimed to identify cancers with weight loss-associated treatment modifications, potential beneficiaries of nutritional intervention. Methods: This cross-sectional study used England’s Cancer Analysis System database, including SACT-treated adults with one tumour and ⩾2 weight recordings between 2014 and 2018. Binary weight loss (threshold: 2.5%) was derived from patients’ most negative weight change from first SACT weight recording. The Martin et al. body mass index-adjusted weight loss grading system (BMI-WLG) was assigned. We describe binary weight loss, BMI-WLG and treatment modification status by cancer. Multivariate logistic regression models of weight loss (binary and BMI-WLG) and a composite outcome of patient treatment-modification status by cancer were produced. Results: Our study population contained 200,536 patients across 18 cancers; 28% experienced binary weight loss during SACT. Weight loss patients were more likely to have multiple types of treatment modifications recorded across all cancers. Regression analyses included 86,991 patients. Binary weight loss was associated ( p < 0.05) with higher likelihood of treatment modification in; colon [Odds Ratio (OR) = 1.72, 95% confidence interval (CI): 1.42, 2.07]; gynaecologic (excl. ovarian) (OR = 1.48, 95% CI: 1.08, 2.01); stomach (OR = 1.6, 95% CI: 1.04, 2.06); lung (OR = 1.38, 95% CI: 1.21, 1.58); leukaemia (OR = 1.30, 95% CI: 1.09, 1.55); head and neck (OR = 1.30, 95% CI: 1.02, 1.65) and oesophageal (OR = 1.29, 95% CI: 1.01, 1.64) cancers. In lung, colon, and grouped gastro-intestinal cancers, association between BMI-WLG and treatment modification increased by WLG. Discussion: Our study is a wide assessment of weight loss during SACT using England’s cancer registry data. Across different cancers we found patients have weight loss-associated treatment modifications during SACT, a precursor to poorer prognoses. Our findings highlight cancers that may benefit from improved nutritional intervention during SACT.