Recent studies have indicated that Large Language Models (LLMs) harbor an inherent understanding of truthfulness, yet often fail to consistently express it and generate false statements. This gap between ''knowing'' and ''telling'' poses a challenge for ensuring the truthfulness of generated content. Inspired by recent work on the practice of encoding human-interpretable concepts linearly within large language models, we treat truthfulness as a specially linearly encoded concept within LLMs, and introduce Adaptive Activation Steering (ACT), a tuning-free method that adaptively shifts LLM's activations in the ''truthful'' direction during inference. ACT addresses diverse categories of hallucinations by utilizing diverse truthfulness-related steering vectors and adjusting the steering intensity adaptively. Applied as an add-on across various models, ACT significantly improves truthfulness in LLaMA (↑142%), LLaMA2 (↑24%), Alpaca (↑36%), Vicuna (↑28%), LLaMA2-Chat (↑19%), and LLaMA3(↑34%). Furthermore, we verify ACT's scalability across larger models (13B, 33B, 65B), underscoring the adaptability of ACT to large-scale language models. Our code is available at https://github.com/tianlwang/ACT.

Despite the great success of large language models (LLMs) in various tasks, they suffer from generating hallucinations. We introduce Truth Forest, a method that enhances truthfulness in LLMs by uncovering hidden truth representations using multi-dimensional orthogonal probes. Specifically, it creates multiple orthogonal bases for modeling truth by incorporating orthogonal constraints into the probes. Moreover, we introduce Random Peek, a systematic technique considering an extended range of positions within the sequence, reducing the gap between discerning and generating truth features in LLMs. By employing this approach, we improved the truthfulness of Llama-2-7B from 40.8% to 74.5% on TruthfulQA. Likewise, significant improvements are observed in fine-tuned models. We conducted a thorough analysis of truth features using probes. Our visualization results show that orthogonal probes capture complementary truth-related features, forming well-defined clusters that reveal the inherent structure of the dataset.

A changing climate is bringing uncertainty to agricultural production and methods that can deliver assessments of crop condition will help mitigate impacts. Agriculture and Agri-Food Canada is developing a vegetation index based on Synthetic Aperture Radar (SAR) to monitor crops. In this research, machine learning algorithms are used to relate SAR parameters from RADARSAT-2 fully polarimetric and Sentinel-1 dual-pol (VV-VH) Single Look Complex data to optical Normalized Difference Vegetation Index values. For four crops (canola, corn, soybeans, wheat) Random Forest Regressors and Least-squares Boosting were able to create strong models using multiple polarimetric parameters. Coefficients of determination (R2) ranged from 0.91 to 0.84 depending on the crop type, sensor and model. Errors for oats and barley were higher due to a more limited training dataset.

Monitoring how farmers till their fields can provide important information in estimating the contributions of the agriculture sector towards soil carbon sequestration and reductions in greenhouse gas emissions. Agriculture and Agri-Food Canada (AAFC) is testing the use of Coherent Change Detection, applied to Sentinel-1 Synthetic Aperture Radar (SAR) data to identify when fields are tilled. Data have been collected in sites in eastern Canada and results to date have been positive. By monitoring the temporal change in coherence, fields that were tilled were successfully flagged using this approach. A more comprehensive data set is currently being collected in order to extend validation of this method and to test if type of tillage can also be identified.

In this study, the dominant and medium scattering phases generated by the Touzi decomposition are investigated for discontinuous permafrost mapping in peatland regions. Polarimetric ALOS2, LIDAR and field data were collected in the middle of August 2014, at the maximum permafrost thaw conditions, over discontinuous permafrost distributed within wooded palsa bogs and peat plateaus near the Namur Lake (Northern Alberta). The ALOS2 image, which was miscellaneously calibrated with antenna cross-talk (-33dB), much higher than the actual ones, is recalibrated. This leads to a reduction of the residual calibration error (down to -43 dB), and permit a significant improvement of the dominant and medium scattering type phase (20°-to-30°) over peatlands underlain by discontinuous permafrost. The Touzi decomposition, Cloude-Pottier α-H incoherent target scattering decomposition, and the HH-VV phase difference are investigated, in addition to the conventional multi-polarization (HH, HV, and VV) channels, for discontinuous permafrost mapping using the recalibrated ALOS2 image. A LiDAR-based permafrost classification developed by Alberta Geological Survey (AGS) is used, in conjunction with the field data collected during the ALOS2 image acquisition, for the validation of the results. It is shown that the dominant and scattering type phases are the only polarimetric parameters which can detect peatland subsurface discontinuous permafrost. The medium scattering type phase, ϕs2, performs better than the dominant scattering type phase, ϕs1, and permits a better detection of subsurface discontinuous permafrost in peatland regions. ϕs2also allows for better discrimination of areas underlain by permafrost from the non-permafrost areas. The medium Huynen maximum polarisation return (m2) and the minimum degree of polarisation (DoP), pmin, can be used to remove the scattering type phase ambiguities that might occur in areas with deep permafrost (more than 50cm depth). The excellent performances of polarimetric PALSAR2 in term of NESZ (-37 dB) permit the demonstration of the very promising L-band long penetration SAR capabilities for enhanced detection and mapping of relatively deep (up to 50 cm) discontinuous permafrost in peatlands regions [1].

Coherent Change Detection (CCD) is applied to exact repeat passes of Synthetic Aperture Radar (SAR) images to identify subtle changes in targets and surfaces. When farmers till their fields, the soil is disturbed and this disturbance can be detected and sometimes measured by SARs. CCD is being investigated as a technique to detect tillage in the Canadian Lake Erie Basin. In this study C-band data from 12 passes of the RADARSAT Constellation Mission (RCM) are processed and compared to field observations. RCM CCD pairs are helpful to distinguish when change happens (due to harvest, tillage and chemical termination of crops). However other scattering parameters, such as volume scattering from the m-chi decomposition, will likely be required to separate harvest from tillage events.

The RADARSAT Constellation Mission (RCM) is able to acquire imagery over large swaths in Compact Polarimetric (CP) modes. The wide area coverage of CP, and revisit with this three satellite constellation, is of potential benefit for operational crop mapping carried out by Agriculture and Agri-Food Canada. This research examined the accuracy of Stokes parameters and m-Chi decomposition parameters, derived from CP data, for identifying crops with a Random Forest (RF) classifier. Stokes S1and S2parameters were plotted on the Poincaré sphere and interpreted as a function of crop phenology. High overall classification accuracies (>90%) were reported when either Stokes vectors or m-Chi decomposition parameters were used in the RF classifier. The Stokes parameters also revealed that the ellipticity, orientation and handedness of scattering varies considerably as crops undergo changes in phenology.

Agriculture and Agri-Food Canada (AAFC) delivers annual maps of crops grown across Canada, operationally, using Synthetic Aperture Radar (SAR) and optical satellite data. This study applies the AAFC methodology to sites in Latin America to test performance and adaptability to these cropping systems, using TerraSAR-X and RADARSAT SAR data. Overall classification results are promising (79.4% to 86.0%), but improvements will occur with better matching of SAR collection dates to local growing seasons, and by acquiring more robust field observations. These improvements will be the subject of additional research by AAFC and partner organizations, in this region.

Due to the characteristics of COVID-19, the epidemic develops rapidly and overwhelms health service systems worldwide. Many patients suffer from life-threatening systemic problems and need to be carefully monitored in ICUs. An intelligent prognosis can help physicians take an early intervention, prevent adverse outcomes, and optimize the medical resource allocation, which is urgently needed, especially in this ongoing global pandemic crisis. However, in the early stage of the epidemic outbreak, the data available for analysis is limited due to the lack of effective diagnostic mechanisms, the rarity of the cases, and privacy concerns. In this paper, we propose a distilled transfer learning framework, which leverages the existing publicly available online Electronic Medical Records to enhance the prognosis for inpatients with emerging infectious diseases. It learns to embed the COVID-19-related medical features based on massive existing EMR data. The transferred parameters are further trained to imitate the teacher model’s representation based on distillation, which embeds the health status more comprehensively on the source dataset. We conduct Length-of-Stay prediction experiments for patients in ICUs on real-world COVID-19 datasets. The experiment results indicate that our proposed model consistently outperforms competitive baseline methods. In order to further verify the scalability of o deal with different clinical tasks on different EMR datasets, we conduct an additional mortality prediction experiment on End-Stage Renal Disease datasets. The extensive experiments demonstrate that an benefit the prognosis for emerging pandemics and other diseases with limited EMR.

Cost-effective permafrost characterization and monitoring should be possible due to advances in the technology of earth observation satellites. In particular, the long-penetration capabilities of L-band ALOS2-PALSAR2 should permit large scale mapping of discontinuous permafrost in peatland areas. Recently, it has been shown that the long penetrating polarimetric L-band ALOS is very promising for boreal and subractic peatland mapping and monitoring [1], [2]. The unique information provided by the Touzi decomposition [3], [4], and the Touzi scattering phase in particular, on peatland subsurface water flow permits enhanced discrimination of bogs from fens; two peat- land classes that can hardly be discriminated using conventional optical remote sensing. In this study, the Touzi scattering phase is investigated for mapping discontinuous permafrost in peatland regions Northern Alberta. Polarimetric ALOS-2 (FP6-4) and field data were collected in August 2014 over discontinuous distributed within wooded palsa bogs and peat plateaus near the Namur Lake (Northern Alberta). The ALOS2 image is re-calibrated to reduce the residual error from -33 dB down to -43 dB. This permits full exploiting the excellent ALOS2 performance in term of low noise floor (NESZ about -38 dB) to increase the sensitivity of the Touzi phase to deep permafrost. It is shown that the information provided by the scattering type phase permits enhanced mapping of discontinuous permafrost. The results obtained with the long penetrating L-band polarimetric PALSAR2 are much better than the ones obtained with conventional discontinuous permafrost mapping methods based on Lidar and optical (Landsat and Spot) images.

ALOS-2, which was launched on the 24thof May 2014, is equipped with a fully polarimeric L-band SAR (PALSAR-2) [1, 2]. Unlike ALOS-PALSAR, which used to collect polarimetric (PLR) data at one incidence angle (about 22°) [3], PALSAR-2 offers the possibility of providing PLR measurements at various beams (FP6-3 to FP6-6), with incidence angle varying from 25° to 35° [4]. Recently, several investigations [5, 6, 8, 9] have been conducted on the assessment and calibration of polarimetric ALOS2, in the context of the ALOS2 calibration-validation (Cal-Val) working group. PALSAR-2 distortion matrix is measured using CRs deployed in the Amazonian forest [5]. The extended Freeman-Van Zyl calibration method introduced in [7] is used for accurate assessment of PALSAR-2 calibration parameters [6]. Six data sets collected over the Amazonian forest (with CRs) are used to assess PALSAR-2 distortion matrix for five beams (FP3 to FP7) with incidence angle varying from 25° to 40°. It is shown that PALSAR2 antenna is highly isolated with low cross-talk (lower than -40 dB) [6]. These results are in agreements with the ones obtained in [8, 9] with different calibration methods.

The availability of advanced satellite radar sensors (C-band RADARSAT-2 and X-band TerraSAR-X) provides significant opportunities for timely monitoring of crop growth. Recent studies revealed that many polarimetric SAR parameters are sensitive to crop Leaf Area Index (LAI). However the reduced swath coverage of fully polarimetric SAR limits the operational application of these modes for large regional monitoring activities. Compact polarimetry mode, on the other hand, permits much larger swath coverage than fully polarimetric SAR. This study investigates the sensitivity of compact polarimetry SAR parameters to crop LAI using simulated data from RADARSAT-2 imagery collected in Canada over two growing seasons. Results revealed that compact polarimetric decomposition parameters associated with volumetric scattering are well correlated with crop LAI. This suggests that compact polarimetric SAR can be an important data source for large scale crop growth monitoring.

This research outlines a preliminary assessment of the use of TerraSAR-X data for classifying agricultural crop land in Canada. X-Band data were able to identify crops (pasture-forage, soybeans, corn and wheat) to accuracies of 95% once a post-classification filter was applied. These accuracies were achieved using six TerraSAR-X images from 2008 and a decision-tree classification algorithm. Acquisitions began only mid-season and consequently a second full season TerraSAR-X data set is being collected in 2009. C-Band classification accuracies were about 10% lower in comparison. These results clearly demonstrate the potential of X-Band data for crop identification.

Agriculture plays an important role in the global economy, and sustainability of this sector is critical for world food security. Annual information on agricultural land use (crop inventory) would permit efficient and effective delivery of agricultural programs that support sustainability of this resource. Previous research has revealed encouraging results on using space borne satellite data (Landsat, SPOT) for crop mapping at the regional scale. Given Canada's large land mass, for operational crop monitoring satellite data with a wide swath and moderate spatial resolution are needed. This study presents the results on integrating RADARSAT-2 ScanSAR data with AWiFS data to improve crop identification. This study demonstrates that multi-temporal AWiFS data can produce an adequate crop classification, with an overall accuracy of 83%. The addition of ScanSAR data increases the overall classification accuracies. The radar contribution is most pronounced during the earlier season.

Mapping and monitoring changes in the distribution of cropland provide information that aids sustainable approaches to agriculture and supports early warning of threats to global and regional food security. This paper tested the capability of Phased Array type L-band Synthetic Aperture Radar (SAR) (PALSAR) multipolarization and polarimetric data for crop classification. L-band results were compared with those achieved with a C-band SAR data set (ASAR and RADARSAT-1), an integrated C- and L-band data set, and a multitemporal optical data set. Using all L-band linear polarizations, corn, soybeans, cereals, and hay-pasture were classified to an overall accuracy of 70%. A more temporally rich C-band data set provided an accuracy of 80%. Larger biomass crops were well classified using the PALSAR data. C-band data were needed to accurately classify low biomass crops. With a multifrequency data set, an overall accuracy of 88.7% was reached, and many individual crops were classified to accuracies better than 90%. These results were competitive with the overall accuracy achieved using three Landsat images (88.0%). L-band parameters derived from three decomposition approaches (Cloude-Pottier, Freeman-Durden, and Krogager) produced superior crop classification accuracies relative to those achieved using the linear polarizations. Using the Krogager decomposition parameters from all three PALSAR acquisitions, an overall accuracy of 77.2% was achieved. The results reported in this paper emphasize the value of polarimetric, as well as multifrequency SAR, data for crop classification. With such a diverse capability, a SAR-only approach to crop classification becomes increasingly viable.

Information on agricultural land use (crop inventory) is needed by various organizations on an annual basis. To meet this operational requirement, Agriculture and Agri-Food Canada (AAFC) has carried out a multi-year (2004 - 2007), multi-sensor (Landsat TM, SPOT, RADARSAT-1, ASAR), and multi-site (five provinces: Ontario, Saskatchewan, Alberta, Manitoba, P.E.I.) research activity to develop a robust methodology to inventory crops across Canada's large and diverse agricultural landscapes. Results clearly demonstrated that multi-temporal satellite data can successfully classify crops for a variety of cropping systems across Canada. Overall accuracies of at least 85% were achieved. When available, multi-temporal (2 to 3 scenes acquired at different growth stages) optical data are ideal for crop classification. However due to cloud and haze interference, good optical data are not always obtainable. A SAR-optical combination offers a good alternative. This research has found that when only one optical image is available, the addition of two ASAR images acquired in VV/VH polarization will provide acceptable accuracies. Of particular interest is the observation that with the incorporation of radar, crop inventories can be delivered earlier in the growing season.